Can we improve our Arm Movement Intermittency with Practice, guided by Auditory and Haptic Feedback? Aaron Shand Creighton
A thesis submitted in fulfillment of the requirements for the degree of Masters of Science in Psychology, The University of Auckland, 2013.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS. Abstract Background: For reasons still not fully understood, people’s movements under constraints, are not smooth movements. That is movements being executed, have fluctuations in velocity. This is true of movements between targets and the slow movements by braille readers. The primary investigation was whether participants could consciously improve smoothness of arm movements with practice. Secondly, can participants meet the requirements of the constant velocity task? Method: The movements were guided by a metronome and haptic feedback from the finger. Participants moved tangentially along two raised lines containing orthogonal bars every one cm, for 18 randomized trials. The lines were 8 cm or 16 cm and movements were executed at one of three driving frequencies 1 Hz, 0.25 Hz and 0.0625 Hz. Intermittency was measured using acceleration zero-crossings and normalized jerk-cost. Results: The task performance data suggests participants could meet the requirements of the task with systematic error. The systematic error suggested haptic distance feedback, auditory time feedback and speed played a role in control of movement. The intermittency data suggests that duration determines the degree of intermittency and the practice effects suggest that we can consciously make movements smoother, but it depends on the level of duration and distance. Post-hoc findings: suggested normalized jerk-cost and acceleration zero-crossings to be sensitive to different components of intermittency and Fast Fourier Transform analysis found three bandwidths, 2 Hz, 3-5 Hz and 7-10 Hz. I discuss the results in terms of theoretical accounts of conscious and unconscious motor control and a link with Parkinson’s disease. Keywords: Cognitive Penetrability, Neurobiomechanical Motor Control, Intermittency, Practice Effects, Haptic Feedback, Auditory Feedback, Perception of Time, Low Frequency Arm Movement and Submovements.
ii
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS. Dedication
Niko Tatopoulos and an unknown gardener
iii
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
iv
Acknowledgements Hello, this Masters thesis could be my only attempt at contributing to science, thus it seems appropriate to attempt to acknowledge everyone who has contributed to my education. I would like to acknowledge my parents, Joarn Shand and Terry Creighton, for guiding me from birth (start location) with feedback and providing a variety of support to enable me to get to this stage in my education (current location). My sister, Amber Shand, for pushing my buttons when we were kids and learning to respect each other as adults. I would like to acknowledge my supervisor Barry Hughes, for allowing me to do this interesting topic and challenging me to write to a high standard. I would like to make a note of the people who have helped me in various ways relating to the technical aspects of this thesis: Josta Heyligers, Julia Novak, Michael Hautus, Hans-Leo Teulings, Jeff Hamm, Carl Helmick, Ian Kirk and Deww Zhang. I would like to thank Garry Venter and Ingrid Ludwig for encouragement in the area of science and critical thinking; my friends Julian Bernard Grau, Eric de la Rosa, Iva Novakov, Jageshwar Sungkur and the garage crew for their support and belief in me; the Reddit community for keeping me ‘sane’ through out the process. It would also be worth noting the support I have received from Lindsay Barron and Laura Seal over the last two years. I would like to mention two previous lecturers (and more if had the space). The first one crystallised my interest in the senses and perception, which led me down this track, Murray White and the second Marc Wilson for his enthusiasm, and the fact that he taught me that I can get A+, if I learn how to express myself efficiently for an exam. I hope to one day emulate their teaching abilities. I would like to acknowledge Bill Watson, who inspired aspects of my interpretation of the study based on the comic strip provided; the team of lecturers in the ABA program who educated me on how to measure behaviour and teach self-control, both of which I applied to myself and was very useful throughout the process; and my participants for their enjoyable questions and participation. I felt very
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
v
honoured to be amongst the students presenting at the 10th Annual InHouse Convention 2013, and a warm thank you is needed for the questions I received, and the ability to be present for some of the presentations. Last but not least, to all my educators past and present, and the scientists and artists who have influenced my education without be ever getting to meet them. These include in order, M. C. Esher, John Nash, Albert Einstein and just recently the work of Richard Fryman that have inspired me to think critically about the world and the beauty of the environment that we inhabit, along with many other artists. To all these people, particularly the people who went out of their way, there is no way I can repay you in kind for the help, and support you have given me. Therefore, I will endeavour to ‘pay it forward’ as I travel through time and space from my current location to the next.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
vi
Table of Contents Abstract..................................................................................................................................... ii Dedication ................................................................................................................................ iii Acknowledgements ..................................................................................................................iv Table of Contents .....................................................................................................................vi List of Figures........................................................................................................................ viii List of Tables ............................................................................................................................ix Introduction ............................................................................................................................... 2 What Causes Movement Intermittency? ........................................................................... 6 Practice Effects on Movement Intermittency .................................................................. 14 Effects of Speed Changes on Movement Intermittency .................................................. 16 Tactile Perception and Haptic Feedback ......................................................................... 18 Rationale for the Present Study ........................................................................................ 22 Method ..................................................................................................................................... 27 Participants ......................................................................................................................... 27 Settings and Apparatus...................................................................................................... 27 Data Processing and Analysis ........................................................................................... 29 Design .................................................................................................................................. 30 Procedure ............................................................................................................................ 31 Kinematic Variables........................................................................................................... 32 Results ...................................................................................................................................... 36 Task Performance .............................................................................................................. 36 Movement Intermittency ................................................................................................... 47
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
vii
Discussion ................................................................................................................................ 60 Task Performance .............................................................................................................. 63 Movement Intermittency ................................................................................................... 70 Acceleration Zero-Crossings or Normalized Jerk-Cost ................................................. 79 Effects of Speed on Intermittency ..................................................................................... 81 Practice Effects and Haptic Feedback.............................................................................. 85 Cognitive Sources of Intermittency .................................................................................. 88 Biomechanical Sources of Intermittency ......................................................................... 93 Applications and Future Research ................................................................................... 99 Perception of Time and Parkinson’s Disease................................................................. 102 Conclusions ....................................................................................................................... 105 Appendix A: Task Performance Data ........................................................................... 108 Appendix B:
Raw Spectrum Analysis Data from MozAlyzeR..................................... 111
References .............................................................................................................................. 113
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
viii
List of Figures Figure 1: Velocity profiles of a person’s finger moving back and forth, at different speeds.. 3 Figure 2: A photo of the apparatus and settings used in the present study…………...…..... 28 Figure 3: Velocity profiles of a participant’s (Kryten) movement under each condition of driving frequency by line length …………………………………...…………...…...…....... 39 Figure 4: Mean distance of segments across trials in each condition (line length by driving frequency).……………………..……….………………………………...……….…...…..... 40 Figure 5: Mean duration of segments across trials in each condition (line length by driving frequency)..…………………….………………………………………...…………...……... 42 Figure 6: Mean speed of segments across trials in each condition (line length by driving frequency). ……………………….……….………………..……………………..…...…..... 43 Figure 7: Mean acceleration zero-crossings as a function of driving frequency for both line lengths.……………….……………………….………….………………..……...…............ 48 Figure 8: Estimates of acceleration zero-crossings across trials (line lengths by driving frequencies). .…………...………………..……...…….….................................................... 49 Figure 9: Mean acceleration zero-crossings as a function of driving frequency for both line lengths...…………...………………………………………………...…..……...…………... 53 Figure 10: Estimates of normalized jerk-cost across trials (segment lengths by driving frequencies)....…………...…………………………………………...…..……...…….......... 54 Figure 11: The relationship between the mean acceleration zero-crossing and mean velocity for the segments is presented....………………………………………...…..……...……….. 57 Figure 12: The relationship between the mean normalized jerk-cost and mean velocity for the segments is presented. ...…………...…………… ……………...…..………….…….... 58 Figure 13: Velocity profiles of two participant’s movements traveling left to right....……. 64 Figure 14: The relationship between the mean number of acceleration zero-crossings and
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
ix
duration for the segments is presented...…………………...…..……...……......................... 72 Figure 15: The relationship between the mean normalized jerk-cost and duration for the segments is presented....…………...…………...…..……...…………………………...….... 73 Figure 16: Velocity profiles with respect to time, of high frequency (1 Hz) long distance (16 cm) conditions for two participants....…………...………………………....……...……....... 76 Figure 17: Examples of velocity profiles for comparing acceleration zero-crossing estimates (AZC) to normalized jerk-cost (NJC) means. ...………………...…..……...…………….... 80 Figure 18: Power Spectrum analysis of the 8 cm high (1 Hz) and medium (0.25 Hz) frequency conditions for three participants...…………...…………………………….…...... 94 Figure 19: Power Spectrum analysis of the 16 cm high (1 Hz) and medium (0.25 Hz) frequency conditions for three participants...…………...……….……...…..…….....…….... 95 Figure 20: Power Spectrum analysis of the 8 cm low frequency (0.0625 Hz) conditions for three participants....…………...………………… ……...…..……...………………….….... 96 Figure 21: Power Spectrum analysis of the 16 cm low frequency (0.0625 Hz) conditions for three participants.....…………...………………….……...…..……...........................…….... 97
List of Tables Table 1: Summary of the Main Task Performance Variables’ Estimates (Standard Error) for the segments............................................................................................................................ 37 Table 2: Summary of Stationary Movement Estimates (Standard Error) for participants’ performance............................................................................................................................. 45 Table 3: The Percentage that Changed in Acceleration Zero-Crossings from the first to the last trials for the six conditions and also a Summary of the Visual Analysis…...................... 52 Table 4: The Percentage that Changed in Normalized Jerk-Cost from the first to the last trials for the six conditions and also a Summary of the Visual Analysis…..................................... 56
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
x
Glossary Afferent: Bearing or conducting something inwards, e.g. nervous impulses towards the brain. Closed-loop: Refers to a type of control system in which the output has an effect on the input in such a way as to maintain the desired output value. A closed-loop system includes some way to measure its output to sense changes so that corrective action can be taken. Cognitive: Of or related to, mental activities involving acquiring and processing information. Cognitively penetrable: A phenomenon which can be modifiable by the person's knowledge or beliefs. Covarying: Two or more variables with correlated variation. Such as strength and sex. Curvilinear: Consisting of, bounded by, or characterized by a curved line. Distal: A point on the body away from the centre. Double dissociation: A general term used of any procedure that enables the ability to distinguish clearly between two functions or processes. Haptics: Relating to the cutaneous senses. Haptics, in the broadest sense, is the study of touch. Note, however, that here I will reserve the term for experiences that come from active touch, touching initiated by an individual. The more inclusive term is Tactile which combines both concepts. Internal model or inverse model: A controller may implement a “inverse transformation” from a combination of feedback and feedforward mechanisms. The feedforward component provides some estimate of the inverse transformation ‘the inverse model’ for a planned movement. During a movement the inverse transformation corresponds to inverting a dynamical transformation that relates an input force to an output motion. Invariant: An entity, quantity, etc., that is unaltered by a particular transformation of coordinates: a point in space, rather than its coordinates, is an invariant. Isometric: Relating to increased muscle activity which does not involve shortening of the
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
xi
muscle and moving the joint. Kinematic: The study of the motion of bodies without reference to mass or force. Merism: Using name of the whole of something, to refer to part of the same thing. The opposite of pars pro toto. Monotonically increasing: A mathematical function that continuously increases. Neurobiomechanics: The range of motion of the nervous system, which results from an interaction of biomechanics and neurology. Orthogonal: Pertaining to objects at right angles (90°) to each other. Paradigm: A particular experimental procedure, e.g. the classical conditioning paradigm. Pars pro toto: Using the name of a part of something (object, event, person etc.) to refer to the whole of the same thing. Phasic: Of, relating to, or of the nature of a phase - having phases. Post-hoc test: Or a posteriori test, is any statistical procedure that is introduced after the data have already been collected and examined. Typically carried out because interesting trends have emerged in the data that call out for examination. Scaling: Scaling is the way that one variable relates to another in similar systems. For example, for similarly shaped objects, the volume and mass scale as the cube of the length, whereas the surface area scales as the square of the length. Shinobi: “Stealth experts” used as secret messengers or information gatherers. Speed: The distance travelled by an object per unit time, regardless of the direction of motion. It is a scalar quantity, which means that it is simply a number indicating a magnitude (of some units, such as meters per second). Steadicams: A trademark for a mechanism for steadying a hand-held camera, consisting of a shock-absorbing arm to which the camera is attached and a harness worn by the camera operator.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
xii
Temporal frequency: The number of times that an event occurs within a given period; rate of recurrence. The period is measured with time, as appose to distance or another form of measurement. Velocity: The rate of change of position with time when the direction of motion is specified. Also known as linear velocity.
Notes: The definitions for the glossary came from a number of sources (Colman, 2001; Credo, 2013; Reber & Reber, 2001; Shadmehr & Mussa-Ivaldi, 1994).
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
CALVIN AND HOBBES © Watterson. Used by permission of Universal Uclick. All rights reserved. This cartoon illustrates the challenges the central nervous system has to overcome when put under constraints, and the unpredictability of behaviour in these situations. It also depicts the desire to improve performance, to win the race.
1
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
2
Introduction A majority of movement in our daily life focuses on moving at a comfortable speed to locations or targets. When aiming for ideal performance many movements, such as during sports, are done as fast and as accurately as possible. However, in certain situations movements are required to be executed slowly and smoothly; for example surgery, painting, Tai Chi or operating a dolly for filming. These situations require putting the movements under constraints. Understanding how the central nervous system (CNS) manages the movements under constraints can provide insights into the control mechanisms of the movement and methods for improving these constrained movements. Understanding how the CNS plans, creates, and controls movement is a key research area in psychology as movements are the fundamental components of our behaviours. One process to understanding the methods used by the CNS can be to determine what factors influence measurable aspects of the movement. For example, when a movement is executed one important measurable dimension is velocity. The velocity profile of a rapid sweep of the hand appears from the naked eye to have a near constant speed. However, when measured, the velocity/speed profile can be described as acceleration of the finger until a maximum level of velocity is achieved (a peak), followed by a deceleration until the movement stops. This profile has been known since Woodworth (1899) but has been elaborated on by research in which high resolution spatial and temporal readings were made (e.g., Atkeson & Hollerbach, 1985; Meyer, Smith, Kornblum, Abrams, & Wright, 1990) and can be seen in examples of fast movements in Figure 1. The next step has been to apply constraints to movement, to see how this profile changes under different conditions. How the profile changes or does not change can be used to infer properties of limb control (e.g., Loram, Gollee, Lakie, & Gawthrop, 2011; Meyer, Abrams, Kornblum, Wright, & Smith, 1988). The results of one of the very
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
3
80
Speed (cm/s)
60
40
20
0
8
9
10
11
12
13
14
15
16
17
18
Finger Postition (cm)
Figure 1. Velocity profiles of a person’s finger moving back and forth, at different speeds.
first studies done by Woodworth (1899) demonstrated that a movement could be corrected while it was being executed. Thereby suggesting a role for feedback during limb movement. This study sparked numerous other studies, some of which discovered a multi-peak velocity profile when a hand approaches a small target (Meyer, et al., 1988; Meyer, et al., 1990), during decreases in speed (Navas & Stark, 1968), and with cyclic movements (Celik, Gu, Deng, & O'Malley, 2009). The increase in the number of peaks during a movement represents an increase in the jerkiness or intermittent nature of the movement. As speed of the movement decreases these multiple peaks appear in the velocity profile (Figure 1). The average person can experience
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
4
this slow jerky movement in day-to-day situations such as, carrying a cup of tea, using a camera or when trying to move like a Shinobi in a water balloon fight. One example of a day-to-day situation is when trying to hold a camera still or film a scene by hand; the slower and smoother the camera is moved, the jerkier the image becomes. Thus numerous devices exist (e.g. dollies, Steadicams, accelerometers or gyroscopes) to compensate for the jerkiness and improve the smoothness of camera movements. Can we achieve such compensation for these movements without additional devices? From the camera example, it is the job of accelerometers or gyroscopes to provide feedback to the device to stabilize the image captured by the camera. With filming, even though a dolly is on wheels or a track (to reduce the jerkiness), it is the job of a person (dolly grip) to make the movements of a dolly as smooth as possible along the desired path in the required timing for the shot. Does a skilled dolly grip compared to unskilled dolly grip have smoother movements or is some other factor involved? Is it even possible for people to have a constant and smooth velocity at various speeds or timing constraints? Does practice enable an increase in performance of these movements? Or is the intermittency of movements, as they become slower, a natural and unalterable consequence of making these movements? This study aims to contribute to the investigation of the CNS to find out if we can consciously compensate for the intermittency and smooth out our movements. The present study investigates aspects of conscious cognition, neurobiomechanics and perception that are related to the fluctuations in constrained movement. Since the discovery of the fluctuations in speed during movements, there has been research aimed at finding what the fluctuations can tell us about the CNS and how it controls movement. These include studies contributing to the understanding of areas such as impaired movement (Hwang, Keates, Langdon, & Clarkson, 2005), neurological control of behaviour (Fishbach, Roy, Bastianen, Miller, & Houk, 2007; Freeman, Dale, & Farmer, 2011), stroke recovery
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
5
(Dipietro, Krebs, Fasoli, Volpe, & Hogan, 2009; Rohrer, et al., 2002), control of prostheses (Doeringer & Hogan, 1995), perception of roughness (Smith, Chapman, Deslandes, Langlais, & Thibodeau, 2002), ageing (Hourcade, 2006; Ma & Trombly, 2004), robotics (Zollo, Salerno, Rossini, & Guglielmelli, 2010), and the area of human-computer interface (Hourcade, 2006). For example, one study by Hofsten (1991; as cited by Rohrer, et al., 2003) found that during development, babies’ movements contain short strait action units (later referred to as submovements). As the infants develop, their movements became smoother as they learned to interact with the environment (or external forces). This finding has been extended with adults with novel external forces (Shadmehr & Mussa-Ivaldi, 1994) with similar findings. Research on the fluctuations in velocity used different terms to describe the multiple peaks in the velocity profile, such as intermittencies (Celik, et al., 2009), submovements (van der Wel, Sternad, & Rosenbaum, 2010), acceleration zero-crossings (Hughes, Van Gemmert, & Stelmach, 2011), and other similar terms (e.g. Slifkin, Vaillancourt, & Newell, 2000). On occasion, the terms ‘submovement’ and ‘intermittencies’ have been used interchangeably (Slifkin, et al., 2000; van der Wel, et al., 2010); however this does not mean that they are synonyms. For the present study these terms are based on the following definitions. Intermittency is used mainly as a simple descriptive term of the movement’s lack of smoothness due to fluctuations in either velocity (Celik, et al., 2009) or higher order derivatives that include acceleration (Meyer, et al., 1988) or jerk (Schneider & Zernicke, 1989). The term submovement implies that each velocity peak is a peak from a single movement (Dounskaia, Wisleder, & Johnson, 2005; Meyer, et al., 1988; Milner, 1992) and implies a specific cause. Number of acceleration zero-crossings is one example of many methods to measure the fluctuations. Zero-crossings of acceleration are a defining characteristic of an intermittent profile indicating the ‘point’ at which acceleration switches
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
6
to deceleration or vice versa (Hughes, et al., 2011). Velocity is often pars pro toto for speed and speed is often a merism for velocity. Within the scope of studying the phenomenon of velocity intermittency the distinction may be important and thus the terms were separated. This study reviews previous research that investigates the ability of the CNS to compensate for the intermittency and smooth out the movements. The review specifically focuses on the interaction between changes in timing and the ability to improve the constancy of velocity of arm movement guided by the finger with practice. It is presumed that the ability to improve the smoothness of movements depends on what creates and controls the multiple peaks in velocity. For example, if the cause was an inherently fundamental aspect of the motor control system, a neurobiomechanical source, then practice and feedback are unlikely to reduce their frequency. Alternatively, the cause could originate from cognitive sources such as feedforward models, sensory feedback or a cognitive execution process. If this is the case then the fluctuations could be cognitively penetrable and a person would be able to consciously reduce the intermittency with practice and feedback. In turn, the effects of practice can contribute to our understanding of the cause of intermittency during movement. The next sections presents the research on common theories of the cause of the multi-peaks profiles, followed by the related research on the effects of practice and changes in speed on intermittency. What Causes Movement Intermittency? Research into the cause of intermittency on changes in movement velocity profiles extends back many years. Within this literature there have been many theories and hypotheses as to their origin (Doeringer & Hogan, 1998; Milner, 1992; Shadmehr & MussaIvaldi, 1994; Slifkin & Newell, 1999). However, a majority of the research since Woodworth’s (1899) seminal work on velocity fluctuations has focused on the cause and control of intermittency during fast movements, such as reaching for a target as rapidly as
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
7
possible. Thus, the theories presented here should be considered within this limitation, since they may not apply to movements of a different type. The five main theories discussed are system noise and intermittent signal (neurobiomechanical sources); and submovement, internal model and closed-loop visual feedback (cognitive sources). Neurobiomechanical sources. Early information models suggested that noise was a part of the biological system (Gawthrop, Loram, Lakie, & Gollee, 2011; Slifkin & Newell, 1999). The system noise theory is that the multi-peaked velocity profile is a result of noise in the system, as the signal is sent from the brain to the muscles to direct the limb or finger. The source of noise in early information system theory was suggested to be the nature of the efferent pathway along which the signal travels (Fitts, 1954; Shannon, 2001). An electric telegram1 can be used as an analogy for this the system noise theory. When sending an electric telegram, relay stations were needed for long distances to keep the signal from degrading, due to the low quality of wire (neural pathway). Recent research has suggested other sources, such as the amount of information (or size of the telegram sent at one time) to increase variability in performance (Slifkin & Newell, 1999). The results of the current studies directly on this theory are mixed. Some studies support this system noise theory (Celik, et al., 2009; Nagasaki, 1991) and other studies do not (Doeringer & Hogan, 1998; Slifkin & Newell, 1999). When investigating the causes of the multi-peak profile of elbow movements, Doeringer and Hogan (1998) demonstrated that the velocity amplitude and noise amplitude were positively correlated, ruling out biomechanical noise, because if noise was the sources of intermittency it should have a negative relationship with velocity, as intermittency does. However, as they mention this did not rule out higher more central sources of noise from the brain (e.g. the telegram was written badly, and parts were missed
1
The electric telegram was chosen for the simplicity of the device. The process of sending an intermittent signal is still used today in various devices such as cell phone (except wirelessly).
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
8
by the operator). Celik, et al. (2009) recorded the frequency of peaks in velocity simultaneously from the shoulder, elbow, wrist and fingertips during circular movements at different velocities. Celik, et al found that velocity fluctuations at these limb locations increased as joint location became more distal during the movements. Based on this finding, Celik, et al suggested that noise was likely a source of the increase in fluctuations since the signal degraded as distance increased. This study had the advantage of having a more natural movement, since they were unconstrained, compared to the restrained arm in Doeringer and Hogan’s study. In addition, Celik, et al were able to simultaneously measure all joints along the arm. Although overall the research conducted has been inconclusive, Celik, et al’s study does suggest that noise acculturating across joints could be a cause of some of the fluctuations in the velocity profile. The second neurobiomechanical theory concerns the method of communication between the brain and the limb. This theory focuses on the cyclic or phasic nature of interneuron signals that are sent from the brain to the muscles (Pasalar, Roitman, & Ebner, 2005). Continuing the analogy, the telegram is sent in bits (or packets of information) with a pause between each bit, word or letter, with a ‘bit’ represented by a period of high frequent neural signal, and a pause having low frequency or vice versa. This has been suggested to be a possible source of the multi-peaks in a velocity profile (DeLong, 1972, 1990; Pearson, 1972). This theory has support from evidence that a phasic signal is sent to the muscles of the wrist (Loram, et al., 2011), finger (Kakuda, Nagaoka, & Wessberg, 1999) and eye (McAuley, Rothwell, & Marsden, 1999). Studies directed at the finger have also proposed a direct connection between a phasic signal and the multi-peaked velocity profile found during certain movements (Flash & Henis, 1991; Milner, 1992). Correlational support has been found between an 8-10 Hz intermittent signal from the finger extensor and flexor muscle on the forearm and the acceleration and deceleration of peaks in the velocity profile of a finger
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
9
(Vallbo & Wessberg, 1993). Groß, et al. (2002) also found a correlation between the oscillatory activity in the cerebello–thalamo–cortical loop and 6-9 Hz velocity fluctuations in finger movements. These two studies suggest that an intermittent signal could be used to control limbs. For the analogy, the brain (or operator) appears to send information in bits at a frequency of 6 to 10 times a second along the neural pathway. If this is the case, an intermittent signal theoretically suggests a form of intermittent control of movement (e.g. Doeringer & Hogan, 1998). Intermittent control has been suggested as a replacement for noise in information models of movement control (Groß, et al., 2002). In addition, intermittent control of movement has been found to be more effective then continuous control for controlling unstable loads (Loram, et al., 2011). Together, the intermittent signal to the muscles and the associated intermittent control model of movement are possible sources of fluctuations in the velocity profile. Cognitive sources. The most common theory of the source of the multi-peaked velocity profile is to do with movement execution and is known as submovements. The submovement concept (stereotypy theory or prototype concept) argues that complex movements are made up of stereotypical submovements, with each submovement being a scalable version of a prototype (Groß, et al., 2002), also termed motor primitive (see, Flash & Hochner, 2005; Giszter, 2009; Mussa–Ivaldi & Bizzi, 2000, for reviews). These submovements are pre-programmed multiple superimposed movements (from one starting posture to another) that replace each other until the movement is complete (Henis, 1991). For a practical example: when controlling velocity of a limb an agonist muscle(s) accelerates the limb, in the desired direction, until the ideal velocity is met or ideal velocity is surpassed. If the limb goes faster than ideal, the antagonist muscle(s) activates and decelerates the limb. If the velocity goes below ideal, the agonist muscle(s) activates and the cycle starts again. The submovements are suggested to be the cause of the fluctuations, because they are not
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
10
completely blended together, creating peaks in the velocity profile (Milner, 1992). Three types of submovements based on a movement profile have been proposed. Two of the types are gross submovements defined by velocity zero-crossings (Type 1) and specific acceleration zero-crossings (Type 2), with the other type being fine submovements (Type 3) defined by specific changes in jerk (e.g. Abrams & Pratt, 1993; Fradet, Lee, & Dounskaia, 2008a; Meyer, et al., 1988). It is worth noting that the submovement theory does not necessarily infer that the submovements are cognitively penetrable, they could be for example initiated by the spinal, basal ganglia or cerebellar systems without the ability to consciously control them. In addition, some types have been shown to be related to biomechanical constraints (Fradet, et al., 2008a: Dounskaia, 2005). However, thought processors could influence the submovements. Development of this theory comes from computational modelling (e.g. Milner, 1992; Thoroughman & Shadmehr, 2000), research with stroke patients (Dipietro, et al., 2009; Rohrer, et al., 2003; Rohrer, et al., 2002) and developmental studies (von Hofsten, 1991; von Hofsten & Rönnqvist, 1993). Hogan and colleagues’ work with stroke patients provides support for the concept of submovements. They found that patients’ movements had distinct submovements, which blended together as they recovered (Dipietro, et al., 2009; Rohrer, et al., 2003; Rohrer, et al., 2002). As mentioned, von Hofsten found blending of submovements as infants developed. Later, von Hofsten and Rönnqvist found neonatal spontaneous movements had distinct characteristics: they changed direction between submovements that straighten out. However not all changes in direction occurred between them. Thus stroke patient and infant research provide evidence that some of the fluctuations are a result of submovements. The next theory is formed on how the brain plans movement in the form of internal models (for reviews see Kawato, 1999; Miall & Wolpert, 1996; Mussa-Ivaldi, 1999), in
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
11
particular a feedforward model (as summarized by Thoroughman & Shadmehr, 2000). This feedforward model suggests that the fluctuations are a result of the planned movement interacting with the environment. This theory suggests the brain builds an internal model of state-dependent approximation of external forces to predict the forces involved in an upcoming movement (Conditt & Mussa-Ivaldi, 1999; Mussa–Ivaldi & Bizzi, 2000). The planned movement follows an ideal trajectory, based on a computational kinematic model. This kinematic model is known as the minimum-jerk model (Ghez, Krakauer, Sainburg, & Ghilardi, 2000; Shadmehr & Mussa-Ivaldi, 1994) which aims for maximization of smoothness (Hogan, 1984). When the planned trajectory is executed, there can be error and the resulting actual movement may not match the computed ideal trajectory (e.g., if exactly the same movement is planned, but in one case it is executed in air and in another case in water, one might expect differences in the actual movement). This can be shown when movement is exposed to a novel environment with novel external forces (Flash & Hogan, 1985). The errors in the trajectory of the movement in these new environments include fluctuations in tangential velocity (Shadmehr & Mussa-Ivaldi, 1994). Together this research suggests a mental or cognitive model, aiming for maximum smoothness, which results in the fluctuations when interacting with the environment. However, the minimum-jerk model for trajectory planning has been criticized (Engelbrecht, 2001; Shadmehr & Mussa-Ivaldi, 1994). A study by Wiegner and Wierzbicka (1992) compared the mathematical model predictions of minimum-jerk and minimum-snap on the resulting jerk-cost and kinematic variables of onejoint goal directed movements. The results supported a minimum-snap kinematic model for fast movements (with no planning for slow movements), which the authors concluded could be accounted for by dynamics of muscle and limbs. Therefore, a feedforward (internal) model, interacting with the environment, could result in the velocity fluctuations obtained, but the research on the minimum-jerk model was inconclusive.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
12
The closed-loop visual feedback theory is based on a combination of concepts to explain the fluctuating velocity profile. Much research within the area of velocity profiles has focussed on the velocity peaks found during fast movements to a target (for reviews see Elliott, 1990; Elliott, Helsen, & Chua, 2001; Meyer, et al., 1990). Out of this research a theory was developed as to the cause of the multiple peaks based on the submovement concept (as described above). The theory interprets the peaks found during movements to a target (e.g. Atkeson & Hollerbach, 1985; Fishbach, et al., 2007) as two submovement types. The first involves an initial primary ballistic submovement which undershoots or overshoots the target due to noise in the system and is then followed by secondary corrective submovements (Meyer, et al., 1988). The secondary submovement peaks that occur are thought to be caused by visual feedback (itself intermittent) correcting for errors in the trajectory when comparing the current effector position against the target position, as the movement gets closer to the target (Crossman & Goodeve, 1983; Elliott, Binsted, & Heath, 1999; Meyer, et al., 1988; Meyer, et al., 1990). A review of the research for and against secondary submovements being visually guided corrective submovements follows. Research on the closed-loop visual feedback theory is not conclusive either. Data in favour of the theory comes from the performance of reaching toward a target (Crossman & Goodeve, 1983; Elliott, et al., 1999; Elliott, Lyons, & Dyson, 1997; Miall, Weir, & Stein, 1993), comparing movements with visual feedback to movements without (Khan & Franks, 2000, 2003; Khan, Franks, & Goodman, 1998), as well as the analysis of the musculature underling the corrective movements (d'Avella, Portone, & Lacquaniti, 2011), the effects of visual perturbations with monkeys (Fishbach, et al., 2007) and research connecting activity in the basal ganglia and cerebellum to submovements (Tunik, Houk, & Grafton, 2009). The study by d'Avella, et al. (2011) provided particularly strong evidence for visually guided corrective movements. Their study on point-to-point movements found that the muscles
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
13
underlying online adjustments to changes in target location are indicated in the secondary peaks, when the target does not move. Data against the theory comes from comparing vision to no-vision conditions (Doeringer & Hogan, 1998; Miall, et al., 1993; Pratt & Abrams, 1996), highly intermittent slow movement during braille reading (Hughes, 2011; Hughes, et al., 2011), and studies on the cause of the types of submovements (Dounskaia, et al., 2005; Fradet, et al., 2008a; Fradet, Lee, & Dounskaia, 2008b; Wisleder & Dounskaia, 2007). Strong evidence against the theory comes from the findings that vision was not needed for intermittency to occur (Doeringer & Hogan, 1998); when it was available, accuracy correction was indicated in only a proportion of submovements (Dounskaia, et al., 2005; Fradet, et al., 2008a; Wisleder & Dounskaia, 2007). Overall the research suggests that during fast movements there are corrective submovements, and visual feedback was one source of submovements. In summary, the research so far has indicated multiple possible causes of intermittency. There is no clear evidence that one theory for the cause provided a better explanation for the evidence then any of the other theories. Some of the sources may contribute to a larger proportion of the intermittency than others, or they could contribute to a type or aspect of the intermittency. In addition, the theories are not mutually exclusive and can be combined. For example, an internal model (influenced by prior learning and occasionally online feedback) sends a signal for each motor primitive (creating an intermittent signal) which is then influenced by system noise (e.g. control of the other joints or signal degrading etc.) on the way to the effector. Thus the literature suggests a combination of both biomechanical sources and cognitive sources being responsible for the existence of intermittency. The effects of practice could indicate the degree to which cognitive sources contribute to intermittency and the research on these effects will be presented first. The ability to
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
14
consciously reduce the fluctuations is also related to movement speed, and this relationship will be discussed in the section that follows. Practice Effects on Movement Intermittency There has been extensive research on the influences of practice during motor skill acquisition (Adams, 1987; Halsband & Lange, 2006; Newell, 1991; Wolpert, Ghahramani, & Flanagan, 2001). The research on the effects of practice on the velocity fluctuations in movement has been growing. This section aims to summarize the research pertaining to the effects of practice on velocity fluctuations. Two examples of research applied to practice effects have already been considered: the feedforward or internal model (Conditt & Mussa-Ivaldi, 1999; Thoroughman & Shadmehr, 2000) and the minimum-jerk model (Hogan, 1984). The tangential fluctuations in velocity that arise in movement trajectories when the internal model interacts with novel external forces can be reduced with practice (Shadmehr & Mussa-Ivaldi, 1994). The minimum-jerk model suggests that when conducting a movement the aim was to maximize smoothness (Hogan, 1984). Smoothness in this model was defined by reduction in the integral of mean squared jerk, the third time derivative of displacement. Research on this model provides evidence that practice increased the smoothness of the movement in an obstacle avoidance task (Flash & Hogan, 1985). However, as mentioned previously, there has been evidence that this model does not apply to fast or slow movements (Wiegner & Wierzbicka, 1992). These models provide possible explanations as to how cognition could play a role during the practice of movements whose velocity is intermittent. There have been studies that found practice-influenced changes in intermittency for the following movement types: fast pointing to target (Khan & Franks, 2000; Pratt & Abrams, 1996; Pratt, Chasteen, & Abrams, 1994), fast arm pointing and wrist rotation (Abrams & Pratt, 1993), and slow movements (Schneider & Zernicke, 1989). The studies on
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
15
pointing to target movements found that the initial and secondary submovements changed with practice (e.g. Pratt & Abrams, 1996). When investigating differences in speed and timing only the slowest movements, when avoiding an object (Schneider & Zernicke, 1989) or grasping (Darling, Cole, & Abbs, 1988), have shown a reduction in intermittency and variability with practice, indicating they can became smoother. However, during a constant velocity task and cycle movements there was no effect of practice regardless of speed (Doeringer & Hogan, 1998). These studies suggest that there can be practice effects, although there may be interdependent factors that determine when they occur. A predominant component of the present study is the relationship between practice effects and feedback. Thus a review of this relationship follows, which focuses on movements without visual feedback. To date the main findings in this area of research indicates the effects of practice to be interdependent with effects of visual feedback (Khan & Franks, 2000) as well as being dependent2 on other factors such as age (Morgan, et al., 1994; Pratt, et al., 1994). The research on the interdependency between practice effects and online visual feedback compared conditions with visual feedback, partial feedback, or no visual feedback. In time-constrained movements visual information has been shown to effect movement control (Proteau & Marteniuk, 1993; Proteau, Marteniuk, Girouard, & Dugas, 1987), and this relationship was dependent on levels of practice (Proteau & Cournoyer, 1990). Visual feedback was also found to improve acquisition of fast and accurate movements, with little change in secondary peaks between the conditions (Elliott, et al., 1999; Elliott, et al., 1997). Following this, improvements in performance, including a decrease in secondary movements, were found regardless of feedback (Pratt & Abrams, 1996). In contrast, Khan and Franks (2003; Khan, et al., 1998) found improvements early on
2
Interdependent indicates that a change in one variable determines the outcome of the second variable and vice versa. Dependent, on the other hand, implies one direction of influence.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
16
in all three types of conditions: vision, partial vision and no vision. However, the vision condition had better performance to start with and the differences increased after extensive practice. They suggested that their results indicate practice with feedback improved the offline programming and use of online feedback. In addition to this, practicing with vision led to improvements in programming and error correction, whereas practicing without vision had generally fewer secondary peaks and these peaks reduced with extensive practice (Khan & Franks, 2000). Khan and Franks (2000) proposed that these differences were due to a progress toward different control strategies. Yet, during a constant velocity and a phase plane task, no effects of practice were found in either type of movement (Doeringer & Hogan, 1998). In general, the research provides evidence that the CNS can influence the intermittency of movements performed, as fast and accurately as possible, without visual feedback and with visual feedback, with this ability being dependent on level of practice similar to time-constrained movements. Effects of Speed Changes on Movement Intermittency The ability to consciously reduce velocity fluctuations is related to the speed of movement. I focus here on changes in intermittency as movement are executed at different speeds, from fast to slow. One early avenue of movement control research that focuses on differences in speed was during target tracking. These studies found an increase in intermittent, jerky movements as velocity decreases (e.g. Navas & Stark, 1968). This finding has been obtained for other types of movements including cyclic movements (Celik, et al., 2009; Doeringer & Hogan, 1998), pointing to a target (Milner & Ijaz, 1990; Morgan, et al., 1994) and a constant velocity task (Doeringer & Hogan, 1998). This negative relationship between velocity and intermittency has been tentatively indicated to be a curvilinear relationship, between 25cm/s and 1cm/s (Celik, et al., 2009; Hughes, et al., 2011). In addition, speed has been found to be the determining variable rather then velocity for
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
17
movement on a vertical plan when compared to on a horizontal plain (Celik, et al., 2009), and when compared to upward against downward movements during a vertical plane object avoidance task (Schneider & Zernicke, 1989). There are movement control theories that incorporate control of voluntary movement as the speed of movements decrease. Two general positions are put forward: control stays the same (Flash & Hogan, 1985; Milner, 1992; Nishikawa, Murray, & Flanders, 1999) or control changes as velocity decreases (Fradet, et al., 2008a; Wiegner & Wierzbicka, 1992; Wisleder & Dounskaia, 2007). According to the submovement theory, control does not change and movements differ only in the number of submovements (Milner, 1992), but more recent research suggests the type of submovements change due to speed and biological constraints (Fradet, et al., 2008a; Wisleder & Dounskaia, 2007). In addition, Wiegner and Wierzbicka (1992) found that neither trajectory model, the minimum-jerk and the minimumsnap, applied to slow movements. Their argument for minimum-jerk model was that the changes in velocity profile did not fit the model proposed by Flash and Hogan (1985) which required the presence of a scaling factor that distinguishes between fast and slow movements. This scaling factor could be the amplitude of 8-10 Hz acceleration cycles. These cycles were found with fourier spectral analysis of slow and fast target-tracking finger movements, and the applitude was the only change detected as speed of the movements decreased (Vallbo & Wessberg, 1993). This idea was also proposed by Terzuolo and Viviani’s invariant rate scaling model with the scaling ratios of submovement during typing (Terzuolo & Viviani, 1980; Viviani & Terzuolo, 1980), hand writing and drawing (Viviani & Terzuolo, 1982). This scaling model has been critizised since it does not apply to other movements (e.g. Gentner, 1987; van der Wel, et al., 2010). However, as Dépeault, Meftah, and Chapman (2008) mention the earlier studies are confounded by the co-variation between duration and velocity. Alternatively, a study by Sittig, Denier, and Gielen (1987) on afferent information
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
18
from the bicep, during non-visually-guided forearm movements, favours the perspective that they are controlled differently. This study indicated that slow moments were dominated by position information and fast movements were dominated by velocity information (Sittig, et al., 1987). Until further research is conducted to find out whether the amplitude of the 8-10 Hz cycles is the only scaling factor, the current research was inconclusive as to whether control changes or does not change. Tactile Perception and Haptic Feedback There are situations in life were haptic feedback may be useful which include: computer-human interface, surgery and braille reading. This study had haptic feedback during movement, instead of the visual feedback used in previous studies (e.g. Wiegner & Wierzbicka, 1992). Thus it would be useful to provide an overview of the tactile perceptual system from surface of the skin to brain. This review focuses on processes involved in the detection of raised elements on a surface and factors that relate to tangential velocity fluctuations (see Chapman, Tremblay, Jiang, Belingard, & Meftah, 2002; Johnson & Hsiao, 1992; Okamoto, Nagano, & Yamada, 2012 for reviews). What distinguishes haptic touch from other types of touch is the active exploration of a surface (Hughes & Jansson, 1994). The tactile system processes five aspects of a surface: its structure, its hardness, its friction (moistness, stickiness) and its temperature (Chen, Shao, Barnes, Childs, & Henson, 2009; Okamoto, et al., 2012). The perception of roughness has been the main focus of research in this area (Okamoto, et al., 2012). Two major issues with researching roughness are, defining ‘roughness’ given it is highly subjective and the different factors that determine a rough surface (Johnson & Hsiao, 1992). For the purpose of this study the perception of roughness will be the ability to detect and process raised elements on a surface. There are four classes of cutaneous mechanoreceptors that can be activated during active contact between skin and a surface: slowly adapting type 1 (SAI) and type 2 (SAII),
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
19
fast adapting RA (FAI) and Pacinian corpuscles (PC or FAII) (Hollins & Bensmaïa, 2007; Wheat, Salo, & Goodwin, 2010). However a recent study by Libouton, Barbier, Berger, Plaghki, and Thonnard (2012) suggests that other receptors not in the finger may be involved. How these receptors process macro and fine roughness has yet to be agreed upon (Chapman, et al., 2002). What the evidence does agree on is that SAI receptors are the critical receptors involved with the processing of raised elements (Gamzu & Ahissar, 2001; Hollins & Risner, 2000; Libouton, et al., 2012; Okamoto, et al., 2012; Yoshioka, Gibb, Dorsch, Hsiao, & Johnson, 2001). The slow adapting and the fast adapting mechanoreceptors are indicated in haptic touch, whereas slowly adapting receptors are used for static touch (Blake, Hsiao, & Johnson, 1997). Both SAI and FAI are indicated in the ability to detect tangential forces (Wheat, et al., 2010). The information from SAI is then processed in the primary somatosensory cortex (Johnson & Hsiao, 1992; Yoshioka, et al., 2001). In particular the temporal frequency of the grating or ridge pattern has been suggested to be represented in the primary somatosensory afferents (Chapman, et al., 2002) and central somatosensory neurons (Darian-Smith & Oke, 1980; Morley & Goodwin, 1987). A key aspect of tactile perception has been that it requires movement of the finger or surface to be effective in processing roughness (Sinclair & Burton, 1991; Sinclair, Pruett, & Burton, 1996). This leads to the question: once the finger is moving does the velocity have an influence? A number of experiments report that changes in velocity have no effect on perception of roughness (e.g. Lamb, 1983; Lederman, 1981; Srinivasan, Whitehouse, & LaMotte, 1990; Yoshioka, Craig, Beck, & Hsiao, 2011). On the other hand, there is research that supports the idea that velocity is important for perception of roughness (e.g. Lederman, 1974, 1983; Meftah, Belingard, & Chapman, 2000). Combining the research, for and against velocity having an effect on perception of roughness, implies temporal cues and spatial cues are used for tactile perception, as other researches have indicated (Cascio & Sathian, 2001;
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
20
Klatzky & Lederman, 1999; Lamb, 1983). These temporal cues can in turn be used for the perception of tangential speed, with spatial cues altering this perception of speed (Dépeault, et al., 2008). With movement along a surface, kinetic friction and force become possibly important variables in the perception of roughness (Cascio & Sathian, 2001; Gamzu & Ahissar, 2001; Smith, et al., 2002). Variations in contact force have resulted in both no effect on roughness perception (Ekman, Hosman, & Lindstrom, 1965) and an effect (Lederman, Howe, Klatzky, & Hamilton, 2004; Loomis, 1985). A study by Lederman and Taylor (1972) found lubrication to have little effect on perception of roughness. In addition, a study by Smith and Scott (1996) found that people were able to modulate tangential forces to detect relative differences in surface texture. With the aim to resolve these mixed results Smith, et al. (2002) investigated the effects of tangential forces, normal forces and their ratio (kinetic friction) on the perception of roughness. Smith, et al found friction was related to perception of roughness, and fluctuation in tangential forces, which are directly related to an intermittent velocity profile, has a significant relationship with raised elements. The frequency of the oscillations increased with mean velocity and spatial period (distance between ridges) and their amplitude increased with spatial period in the longitudinal scanning direction. In their second experiment, Smith, et al provided evidence that the fluctuation in tangential force may be a determinate of roughness perception. In accordance with this, aspects of element do have an effect on perception of roughness (Blake, Hsiao, et al., 1997; Chapman, et al., 2002) and the frequency of acceleration zero crossings has a relationship with mean velocity (Blake, Johnson, & Hsiao, 1997; Cascio & Sathian, 2001), but frequency of acceleration zero crossings does not increase with an increase in the density of dots in a braille cell (Hughes, et al., 2011). Thus the research suggests that friction and force affect the perception of roughness, but their relationship with velocity fluctuations remains unclear.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
21
To summarize, when a finger is moving along a grating pattern, a number of factors are involved: receptors, movement, friction and tangential forces. The main receptors that detect the raised element on the surface are SAI; these interact with FAI to detect tangential forces. Movement is required which enables both temporal and spatial cues to be detected and processed and tentatively suggests that the finger pad, depending on whether the finger is moving or the surface is moving could perceive a surface differently. During the movement, the friction with the surface influences the person’s perception of roughness. The velocity of the movement and the roughness of the surface can influence fluctuations in tangential forces. These tangential forces can be consciously modulated to influence the perception of roughness. This suggests that fluctuations in tangential forces can influence cognitive decision-making. The tactile perception process and the haptic feedback are a component of this study. These components should be combined with the previous presented research, in order to acknowledge the tactile perceptual factors within this study. The majority of research across the different areas of psychology and other disciplines on this topic of fluctuations has focused on visually guided fast movements to a target a number of degrees away from a start location (e.g. Adams, 1987). Some studies have removed visual feedback (Pratt & Abrams, 1996), some studies have used other types of movement (e.g. cyclic, Celik, et al., 2009), some studies have restricted duration and distance (Morgan, et al., 1994) or had constant velocity tasks (Doeringer & Hogan, 1998), and some studies have looked at practice effects (Khan & Franks, 2000) or surface texture (Smith, et al., 2002). This research has yet to establish if people can consciously reduce the intermittency of their movements, in particularly very slow movements, with haptic feedback. The present study adds to the literature by having a unique combination of environmental stimuli to fill the gap in the literature. The investigation was to determine if people can consciously improve the smoothness of arm movements guided by the finger
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
22
along a raised line (haptic feedback), while blindfolded. The requirements for the movements were constant velocity under different timing and distance constraints. The distance was fixed in centimetres as opposed to degrees, as some tasks in day-to-day life are not relative in size to the person, such as moving across a page width of braille cells. Participants had no visual information of the task surface provided at all. Due to the novelty of these combined conditions, the present research also focuses on peoples’ ability to move as required, and the effects of the constraints and practice on the fluctuations. The reasoning for the study will be explained next. Rationale for the Present Study This study aims to establish whether people consciously improve with practice the smoothness of movements, guided by haptic feedback. In doing so this could provide an indicator to the level of contribution to intermittency, from cognitive sources or biomechanical sources. In order to pursue this aim, participants were blindfolded and given repeated opportunities to move at a single finger along a raised line (in the left-right dimension) at “as constant and as smooth a velocity as possible” under three ideal cyclic timings (driving frequencies) over two ideal distances (line lengths). These conditions would enable the investigation of a range of ideal speeds of movement, from fast (32 cm/s) to very slow (1 cm/s). In practical terms, this procedure was similar to asking a dolly grip to move the dolly from a start location to a target, in three fixed times and then repeat this with two distances, along the desired path for the camera. Additional online haptic feedback was given to enable people, if they could, to determine their movement velocity. To explore this behaviour, kinematic variables of the movements were measured at the tip of the finger, and these variables included velocity, acceleration and jerk. This study was a systematic replication of an earlier unpublished report by Mathur
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
23
and Hughes (2009) with the addition of haptic feedback and other minor changes3. Both studies originated from braille research by Hughes (2011; Hughes, et al., 2011). Hughes and colleagues found, contrary to claims that reading velocities are constant and smooth, highly intermittent velocity profiles during braille reading. Further, the intermittency was related to mean reading velocity, word frequency and difficulty level (as reading became easier during a second reading it also became faster and less intermittent). Mathur and Hughes (2009) extended previous research by finding no effects of distance and an inverse exponential relationship between velocities (over a range of 1-45 cm/s) and intermittency, with no visual feedback. Mathur and Hughes had participants make finger movements along a flat surface and raised horizontal line, under three driving frequencies (high, 1 Hz, medium, 0.25 Hz and low 0.0625 Hz) and two line lengths (8 cm, 16 cm). These constraints were duplicated for this current study, with one variation: haptic information was made available. This information could, in principle, provide additional velocity information that could then be used to improve the smoothness of their movements. The additional haptic information consisted of short raised bars orthogonal to the direction of motion, at 1 cm intervals along the horizontal line. This manipulation was expected to indicate whether movements deviated from a constant velocity. The reasoning for this was as follows: as participants moved their finger pads from start to finish, the frequency at which participants encountered the raised bars had been shown to inform participants of the nature of the speed at which they were traveling (Dépeault, et al., 2008). If participants’ experience a fixed or stable temporal frequency this would indicate a constant velocity, whereas a variable temporal frequency would indicate a deviation and participants could then adapt the
3
These were primarily small methodological changes and included for example; no visual information of the surface for participants, trail order was accurate, a different method for dealing with trials that had error and measurement of dwell time, stop periods etc.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
24
movements. In other words, if tangential fluctuations in force4 can be used to modulate the perception of roughness (Smith, et al., 2002; Smith & Scott, 1996), then can the reverse occur to control fluctuation in velocity and force? Hypotheses. The design of the experiment enables the test of several hypotheses. These were split in two goals to determine if participants can move continuously at a constant velocity under the different driving frequencies and line lengths, and to determine the influences of the factors duration, distance and practice on intermittency. It was hypothesized that participants could move as required under the different conditions of the study. Evidence suggesting this hypothesis would come from their performance matching the expected performance. The expected performance was based on the ideal performance with systematic performance distortions. These systematic distortions would occur during their actual movements, and would be distributed around the ideal performance. Additional systematic performance distortions would be in the form of movements having higher than ideal speed and overshoots which were found by Mathur and Hughes (2009). The overshoots would likely be predominant during the high driving frequency conditions, as they require fast movements and are likely to be ballistic in nature (Atkeson & Hollerbach, 1985). Minimal duration of stationary movement, such as dwell time was also expected (Morgan, et al., 1994). Effects of practice for these variables were expected, as the participants are likely to be able to improve these aspects of their performance (Adams, 1987; Halsband & Lange, 2006; Newell, 1991; Wolpert, et al., 2001). The practice effect would particularly occur in conditions that are difficult for the participants and the practice would then enable them to meet the requirements. Distance under fixed driving frequencies would have no effect on intermittency as
4
Fluctuations in force are directly related to fluctuations velocity because of their relationship with mass and mass would be fixed for each person’s finger/arm in the studies by Smith and colleagues.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
25
found by Mathur and Hughes (2009). As line length increases there would be more opportunity for fluctuations to occur, if duration was proportional to distance travelled. However, since duration was fixed, then there would be the same opportunity for fluctuations to occur. This would indicate that the shorter movement was a more compact version of longer movement, which was similar to the invariant rate scaling model (Terzuolo & Viviani, 1980). Evidence for this hypothesis would be from finding no main effect of distance on intermittency. Duration, measured by driving frequency, of movement intermittency was hypothesized to follow a negative curvilinear relationship. This was based on driving frequency and intermittency having a similar relationship as speed does with intermittency (Celik, et al., 2009; Hughes, et al., 2011; Mathur & Hughes, 2009) since an increase in driving frequency for a fixed distance determines an increase in ideal mean speed. Evidence for this hypothesis would be from a significant negative curvilinear trend across driving frequencies. This was then further investigated in respect to velocity. The additional investigation was to determine if an inverse exponential function would fit the relationship between intermittency and velocity, similar to Hughes, et al. (2011; Mathur & Hughes, 2009). It was hypothesized that practice effects would occur. This was based on the extensive findings of the effects of repeating behaviour that results in an improvement. Practice effects without online visual feedback of the hand in movements to a target (Khan & Franks, 2003; Khan, et al., 1998) and time constrained movement (Proteau, et al., 1987), maybe used as a basis for this hypothesis. Evidence for this hypothesis, would be a significant negative linear trend across trials, from polynomial contrast effects. If practice effects occur, the hypothesised relationship between distance and practice was that, as line length increases the ability to improve the smoothness of the longer
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
26
movements would increase, compared to the shorter movements. If the hypothesis between distance and intermittency holds, then the short movements are a restricted version of the longer, faster movements. Thus it would be easier to improve the less restricted movement. Evidence for this hypothesis would be either a practice effect only for the long line length (16 cm) conditions or a greater reduction in intermittency during the longer line condition. If practice effects occur, the hypothesised relationship between duration and practice was that movements would become smoother at the medium and low driving frequencies conditions. This hypothesis was based on practice effects only being found during visually guided movements (Darling, et al., 1988; Schneider & Zernicke, 1989) that had a similar velocity as the medium frequency conditions in this study. Evidence for this hypothesis would be a decreasing trend in intermittency across trials in the two conditions. The decreasing trend across trials may or may not level out after a number of trials. If the high frequency movements do become smoother with practice these would be limited, due to floor effects, since the high frequency movements are faster and thus typically have few peaks in velocity, with a minimum of one peak. The outcome of the test for practice effects could spread to the theories of the possible sources of the velocity fluctuations. A reduction in intermittency due to practice effects would support the idea that some of the fluctuations could be cognitively penetrable and favour a cognitive source. If practice effects do not occur, then the intermittency is not cognitively penetrable, which would favour neurobiomechanical sources. If the intermittency levels out after a period of practice, then the majority of the remaining fluctuations are likely the result of neurobiomechanical sources. The methodology and results of this study could also be useful for designing real world techniques. These techniques can be used for improving the performance of situations that require smooth movements, which include the situations mentioned painting, surgery or operating a dolly.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
27
Method Participants There were 18 participants without perceptual, cognitive or motor impairments who gave informed consent under protocols approved by the University of Auckland Human Participants Ethics Committee. The participants (11 female, 7 male) ranged in age from 18 to 43 with a mean age of 22.94. They where asked to use their preferred finger for the task. 14 were right-handed and used their right index finger. Four were ambidextrous and half of these four used their left index finger. One was right handed and used his right middle finger. For their participation they received a $30 voucher. Settings and Apparatus Surface. The participants sat at a table, with the middle of the tablet aligned with their shoulder and their finger resting on the surface (Figure 2). This figure depicts the surface, which was one of two plastic sheets, which were embossed with raised lines. The plastic sheets were pliable relative to other studies (Smith, et al., 2002), and this made it difficult to measure with consistency the height of the elements. However, this did enable the ability to adjust the height of the elements for each participant. This was done prior to the experiment by asking and observing if participants feel the elements of the lines and bars. The surface was on a plane of either (x= -3.9°, y=1.6°) or (x=-5°, y=2°), which has a slope toward the participant and toward the participant’s right. In absolute and practical terms this slope is minimal and was considered horizontal for the purposes of the study. Both sheets had two horizontal lines on the sheet, one 8 cm and the other 16 cm. Both lines had 3 cm vertical lines at their ends, with 1cm being above the line. Between the ends of the two horizontal lines, 1cm long vertical bars were placed with a spatial period of 1 cm (Figure 2). Both horizontal lines were aligned to the left, with one sheet having the 16 cm above the 8 cm line and the other sheet having the 8 cm above. The sheets were split evenly between
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
28
participants in a random order.
Figure 2. A photo of the apparatus and settings used in the present study. The raised lines on the sheets were 16cm and 8 cm long, with vertical bars at 1cm special period. The digitizer’s pentip with the lightweight device are attached to the finger, with the electronics under the sleeve. The rubber was used to inform them, were to put their finger after a trial (i.e. not on the recording surface), which also worked as a reference point: as to where the lines were in space. The white paper was added to the scene so the raised lines would be easily visible to the reader.
Recording device. Movement of the finger was recorded on a 30.5 cm by 30.5 cm digitizing tablet (Intuos2 by Wacom), with a 0.005 mm resolution and a clock speed of 200 Hz. The digitizer’s pentip was fitted to a lightweight finger attachment, so that the pentip sat
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
29
in a fixed position at the end of the finger (Figure 2). The electronic componentry for the pen was relocated to the participant’s forearm and was secured with a compression sleeve. The pentip was in a near vertical relationship to the finger. This arrangement was the similar to that done by Hughes (2011; Hughes, et al., 2011). The movements of the pentip were sampled at 100 Hz using the tablet and MovAlyzeR (NeuroScript, 2006-2010). Timing device. A digital metronome (Fleckner, 2010), was used to guide the driving frequencies of movements. During the instructions it was turned on, and appropriate volume of the device was checked for each participant. The metronome continued until the end of the experiment sounding at 2 Hz (120 bpm). This metronome was used (contrary to some anecdotal reports) because no errors were noticed in the timing of beats during an hour pretest phase or during the experiment. Data Processing and Analysis The position data were then differentiated with respect to distance and sampling period for velocity, acceleration and jerk using MovAlyzeR with a 4th-order-zero-lag Butterworth low-pass filter method (at a frequency of 12 Hz). Post processing of this data involved the cropping of any data recorded before the initiation of movements for the trial and any excess information recorded after completion of the required movements. A start or stop of the entire movement was defined by as velocity zero-crossing before or after the main proportion of the movement (represented by high levels of velocity relative to the condition). However, some trials did not fit this definition well, some trials had extensive periods before or after with velocity relatively close to zero. Some trials had movements finished, which did not get close to zero. In these situations the valley closest to zero after the main movement was chosen. There were other rare situations that did not fit the situations mentioned, which were measured with a similar approach and the same aim.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
30
The data was processed for the kinematic variables in relation to the x dimension (tangential left-right) in Microsoft’s Excel (2010). The kinematic variables related to task performance were calculated for means and standard error. The main kinematic were analysed using a factorial repeated measures analysis of variance (ANOVA). The intermittency measures were statistically analysed with priori tests of ANOVAs and polynomial contrasts (with adjusted spacing for driving frequency) using IBM’s SPSS (v. 20). A α of 0.05 was chosen for the analysis. Systat Software’s SigmaPlot (v.11) was used for fitting lines for the figures. During post processing, if an error in a trial was noticed (that was missed during the experiment) then these trials became an average of the previous trial and the next trial. There were a total of 20 trials (1.03%) that were missed. Design The design of study was the same as Mathur and Hughes (2009) and grounded on a 2x3 design across randomized trials, with two line lengths and three diving frequencies. The line length conditions involved movements along a horizontal either 8 cm or 16 cm line to the end and back. The three driving frequencies conditions were high 1 Hz, medium 0.25 Hz and low 0.0625 Hz. These cyclic movements are often described as tangential sinusoidal movements. The 1 Hz conditions, required participants to complete, eight cycles of moving to the end of the line in one beat (.5 s) and back in one beat. The 0.25 Hz conditions required participants to complete, two cycles of four beats, 2 s per direction. The 0.0625 Hz conditions required participants to complete, one cycle of moving to the end of the line in 16 beats, 8 s per direction. Each trial was aimed at having the same duration. However, due to the increase in duration a low frequency movement would have in order to make a complete cycle, this had double the duration of the other two.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
31
Procedure The procedure was in the following order; instructions, exploratory phase, practice phase and the experiment. Participants’ were given five basic instructions. First, they were informed that they would be required to “move their preferred finger along a short, 8 cm line, and a long, 16 cm line, under three different speeds”. Secondly and primary request was, “to attempt to move as smooth and as constant a velocity as possible, in the specified time”. This primary request was repeated in different ways throughout the instructions. Thirdly, they were told that there were short vertical lines at 1 cm intervals so that “if it feels like you’re encountering them at a constant frequency, then you are traveling at a relatively constant speed”. The forth instruction was that they were asked to reduce the time they stopped at the end of the line (what we refer to as dwell time). This was aimed to prevent the possibility that they would dwell for long periods (van der Wel, et al., 2010). Lastly, they were instructed, “if you are traveling at the wrong speed, just stay at the same speed and keep it smooth until you get to the end of the line, then correct it for the next repetition”. This instruction was to emphasize the primary focus of smooth movement, and secondary focus on accuracy. At the end of the instructions, participants were blind folded and remained blindfolded until the end of the experiment. At the start of the exploratory phase the tablet was uncovered. In this phase each participant’s finger was placed at the start of both lines and given a chance to explore the length of the lines. An important part of this phase was to ask them if they could detect the raised elements, of the lines. If they could not, the elements were raised. The practice phase contained two of each trials of each condition (i.e., 12 trials in random order). These trials were conducted to make sure the participants understood the movements required for each trial. Verbal feedback from the experimenter was given, if their movements did not match what was required for each trial. If they reversed direction before
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
32
the end of the line at any point (but mainly in the low driving frequency conditions) then the practice phase was paused they were asked to move to the end of the line without listening to the metronome. If they could not do this, the parts of the line they did not move over were raised. This was repeated (but rarely) until they could move to the end of the line, after this the practice phase was continued. The experiment contained 18 repetitions of each of the six unique conditions, resulting in 108 randomly ordered total trials. The metronome was audible continually and participants were instructed to start moving when they were ready. This was done to enable them time to plan the movement if necessary, and to have the movements be self-initiated. The body was not restrained in any way, this included no leaning (of the hand or pentip) on the tablet (Figure 2) thus the joints involved were numerous. The predominant joints involved appeared to be the elbow, followed by wrist, finger and shoulder. If an obvious recording error or external disruption occurred during a trial such as a loud noise, then the trial was relocated to the end of the trial sequence and conducted again when the time came. Kinematic Variables The kinematic variables were measured at the tip of the finger during the movements. Unless otherwise stated, all the kinematic variables were processed per segment and then averaged for the trial. A segment was defined as a movement going either left or right, including regressions or stationary movement (stop periods) and excluding dwell periods. A dwell period was stationary movement or stop during a change of direction between segments. Dwell periods, stop periods and regressions are defined further below. Task Performance. The main task performance variables measured the change in the pentip’s location through time and space. Line length measured the ideal distance of a segment before changing direction. The actual movement was measured as distance (cm) travelled for a segment. Driving frequency defines ideal duration of a movement, from one
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
33
location to another location, with respect to one second. Since the movements are cyclic the start and finish location are the same. The actual duration of movement performed was measured in seconds, for each segment. As mentioned driving frequency and the line length defines the ideal speed5 for each segment. The actual speed and velocity during a movement was measured by taking the mean of the sampled values of absolute velocity (cm/s) for a whole trial, excluding dwell periods. Practice effects for both these variables and the intermittency variables were defined as: a decrease (or increase if ideal performance is above start performance) in the trend of the mean across trials, which may or may not level out. Five additional variables were processed to measure expected movement that deviates from the ideal performance. The ideal performance was defined as: to move continuously for the whole trial in the correct direction. Dwell periods (percentage of occurrence) and stop periods (percentage of trials and count per segment) are measures of the quantity of stationary behaviour. Dwell time/stop time, were a measure of average duration in seconds of the associated stationary movement. Regressions (n per condition) are a measure of movements in the opposite direction. The percentage of dwell periods was calculated by taking the number of occurrences divided by the number of opportunities for it to occur. Dwell time and stop time were calculated the same way, by taking the average duration of each of the periods when they occurred. The percentage of trials with stops was calculated by taking the number of trials with stops and dividing by the total number of trials. Stop periods was calculated by taking the average number of stops per segment in each trial divided by the number of trials that had stop periods. All five measures were calculated for each subject (per condition) and then each subjects value or percentage averaged for the total condition.
5
Driving frequency determines ideal duration, line length determines ideal distance and both determine ideal speed. The speed and direction of the line determines ideal velocity.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
34
The characteristics of dwell periods, stops, and regressions were based on analysis of velocity profiles from the study by Mathur and Hughes (2009). A dwell period had four characteristics: 1) Velocity was within a window of less then or equal to ±0.26 cm/s. 2) Occurred for at least 0.04 s. This was used for a couple of reasons, the primary reason being that when velocity on the x dimension reaches zero, it does not mean that the participants’ finger was ‘stationary’. An example of this concept can be found in the study by Fradet, et al. (2008a). 3) If velocity of the movement left the window for a period, this was included if less then 0.07 s. This was included since participants are ‘stationary’ but the pentip is responding to isometric intermittency (Slifkin, et al., 2000). 4) Occurred during a change in direction at the end of a segment (velocity zero-crossing). A stop was defined the same as dwell period, except that it happened during a segment and velocity had to reach zero at least once. A regression was calculated as a period where movement went the opposite direction for more then 0.04 s during a segment. Intermittency. This was measured using two measures: the number of acceleration zero-crossings and normalized jerk-cost. These measure the ability to execute a constant and smooth velocity. The number of zero-crossings of acceleration indicates the number of times the velocity changed, from acceleration to deceleration or vice versa. The larger the number of zero-crossings, the more fluctuations (peaks and valleys in the velocity trace) there were. Acceleration zero-crossings were used for a number of reasons, including intuitive simplicity and the ability to compared to studies with similar methods (Celik, et al., 2009; Hughes, et al., 2011). Normalized jerk-cost is a dimensionless version of jerk-cost, the integral of squared jerk (the third motion derivative). This measure indicates the degree to which a movement deviates from a smooth profile, represented by jerk-cost value of 7.57 (Teulings, ContrerasVidal, Stelmach, & Adler, 1997). The formula for this measure was adapted from Teulings,
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
35
et al. (1997). The addition of the square root enables normalized jerk to be proportional to absolute jerk (Teulings, et al., 1997), and creates simpler units from the slower movements. The Riemann sums were used as an approximation of the integral, since the measure was discrete (Anton, Bivens, & Davis, 2005). The normalized jerk-cost formula was calculated for each segment and follows: 1 2
!
𝑗 ! 𝑡 0.01 × Duration! /Length! !!!
Normalized jerk-cost was used as a measure of intermittency for a number of reasons. The primary reason was that normalized jerk-cost could have a large difference in what aspects of intermittency were measured, when compared to acceleration zero-crossings. This was suggested based on the findings by Rohrer, et al. (2002), who found jerk-cost (normalized by peak velocity) to measure different aspects of intermittency than peak count. The version of normalization of jerk-cost used here was chosen because it has been promoted as an ideal measure of intermittency and one of the best versions of jerk-cost measurement (Hogan & Sternad, 2009). Hogan and Sternad (2009) suggest this was because it aims to enable the ability: to compare across studies, to compare movements of different duration and length, and to account for stop periods, peaks and valleys (fluctuations that do not result in a stop). These advantages, specifically the ability to account for stops, could make this measurement more sensitive than acceleration zero-crossings. The ability to compare movements of different duration and length is due to the normalization process, and it enables the jerk-cost formula (designed for a fast as possible trajectory) to be used for a constant velocity task and to directly compare the two tasks (Hogan & Sternad, 2009; Teulings, et al., 1997). This benefit emerges because it controls for the differences in duration and distance between these movements.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
36
Results The results are split in to two sections, one for each goal. The first section focuses on if participants could perform as expected based on the previous literature. The second section focuses on the influences of the independent variables driving frequency, line length and trials on intermittency. Task Performance This section of the results will present a description of the participants’ movements under the different timing (duration) and distance constraints of the task. This descriptive process starts with a visual analysis of example velocity profiles. Then the main task performance variables will be investigated, with the full 2 x 3 x 18 (Line Length [8 cm, 16 cm] x Driving Frequency [0.0625 Hz, 0.25 Hz, 1 Hz] x Trial [1 to 18]) analysis of variance model ANOVA. The main task performance variables are presented first with the estimates for each condition and then the ANOVAs. Followed by the five additional variables, measuring stationary movement and regressions for each condition. The estimate variables measuring the actual movements for each condition are in Table 1. These can be used to indicate if the participants’ movements’ distance (longer then line length), duration (driving frequency) and speed (higher then ideal speed) were similar to the associated expected performance, over the whole conditions. The average distance travelled by participants was longer then the line lengths as was expected and the overshoots were around 1 cm to around 2.5 cm depending on condition. The overshoots increase with driving frequency and there were small change between line lengths. The results for duration in the 16 cm segments indicated duration was as required, with a distribution around the ideal performance. However, in the 8 cm distance the duration was shorter then required for all three driving frequencies. The speed of movements was higher as expected and increased with line length and driving frequency as expected.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
37
Table 1 Summary of the Main Task Performance Variables’ Estimates (Standard Error) for the segments. The six conditions are Line Length (8 cm, 16 cm) by Driving Frequency (high 1 Hz, medium 0.25 Hz and low 0.065 Hz). Distance (cm) Conditions
Duration (s)
Speed* (cm/s)
M (S.E.)
8 cm 1 Hz
10.60 (0.25)
0.47 (0.01)
22.73 (0.57)
0.25 Hz
9.58 (0.13)
1.88 (0.02)
5.14 (0.10)
0.0625 Hz
9.03 (0.12)
7.49 (0.08)
1.23 (0.02)
1 Hz
18.42 (0.29)
0.51 (0.01)
36.57 (0.64)
0.25 Hz
17.91 (0.17)
2.01 (0.03)
9.00 (0.14)
0.0625 Hz
17.27 (0.12)
8.05 (0.10)
2.16 (0.03)
16 cm
*Speed was measured by taking the average of the online speed during the segments
Velocity profiles of a single participant’s (Kryten6) movement during each condition are presented in Figure 3, visual analysis follows. The effect of distance appears to generally only increase the velocity of the movement as would be expected under the conditions. The ability to move continuously appears to be mainly disrupted by a low driving frequency, indicates that moving continuously became harder. Changes in driving frequency also appear to impact the ability to maintain constant velocity. Maintaining a constant ideal velocity would look like an increase in velocity until the ideal velocity was met and this would be ‘
6
This is a pseudo-name, or fake name, for the purposes of a small n-design study conducted in the discussion.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
Line Length: 8 cm
Line Length: 16 cm 15
15
Driving frequency: 0.0625
Driving frequency: 0.0625
10
10
5
5
0
0
-5
-5
-10
-10
-15
-15 4
6
8
10
12
14
16
18
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28
Finger Position (cm)
Finger Position (cm)
30
Finger Velocity (cm/s)
38
30 Driving frequency: 0.250
Driving frequency: 0.250
20
20
10
10
0
0
-10
-10
-20
-20
-30
-30 4
6
8
10
12
14
16
18
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28
Finger Position (cm) 90
Finger Position (cm) 90
Driving frequency: 1.000
60
60
30
30
0
0
-30
-30
-60
-60
-90
-90 4
6
8
10
12
14
Finger Position (cm)
16
18
Driving frequency: 1.000
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28
Finger Position (cm)
Figure 3. Velocity profiles of a participant’s (Kryten) movement under each condition of driving frequency by line length. The driving frequencies for the cyclic movements were high (1 Hz), medium (0.25 Hz) and low (0.0625 Hz). The thin lines are the velocity profile of the first trial of each condition and the thicker lines are the last trial. The grey box represents the location of the horizontal raised line with vertical lines at 1 cm. Note: the ordinal scales vary according to segment length and driving frequency.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
39
maintained and then a decrease at the end of the line. Overall, the medium and low frequency conditions appear to follow this pattern, with fluctuations around a mean velocity. The high frequency conditions appear to follow the pattern, however the movements are ballistic as was expected. This was indicated by, the velocity continuing to increase above the mean velocity until it reaches a relatively flat peak and then decreases. In addition, the profiles depict the expected overshoots of the ends of the lines, where this participant changed direction after the end of the line. Overall, the profiles indicate that Kryten’s performance matched expected performance. Distance. All main effects were significant, line length F(1, 17)=4539.58, p<.001, MSE=7.08, driving frequency F(2, 34)=39.84, p<.001, MSE=7.56 and trial F(17, 289)=3.65, p<.001 MSE=7.44. One significant interaction was found between line lengths and driving frequency F(2,34)=11.61, p=.014 MSE=2.39. No other interactions were significant. These results do not support the idea that, line length alone determines distance travelled. However, it was expected that there would be overshoots over the end of the line, and that these would be determined by a change in driving frequency (Table 1). The effects of trial suggest possible effect of practice, and this appears to be the cases in some of the conditions were overshoots actually increased, which was not expected (Figure 4). In addition, Figure 4 indicates that the overshoots occurred in all conditions and increase with driving frequency, as expected. Duration. It’s expected two significant main effects were found, line length F(1, 17)=49.82, p<.001, MSE=0.58 and driving frequency F(2, 34)=9279.79, p<.001, MSE=1.04, trail was not significant F(17, 289)=0.49, p=.956 MSE=7.44. One significant interaction between line length and driving frequency F(2,34)=22.69, p<.001, MSE=0.55 was found. The other interactions were non-significant. These results do not support the idea that driving frequency alone determines duration of movement. The expected duration of a segment
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
LIne Length: 8 cm
Line Length: 16 cm 18
10
Driving frequency: 0.0625
Driving frequency: 0.0625
17
9
y = 8.892 - 0.014x r ² = 0.197
y = 17.204 + 0.007x r ² = 0.06 16
8
1 2 3 4 5 6 7 8 9 101112131415161718
1 2 3 4 5 6 7 8 9 101112131415161718
Mean Distance (cm)
11
Trial
Trial 19
Driving frequency: 0.250
Driving frequency: 0.250
10
18
9
17
y = 17.723+ 0.002x r ² = 0.414
y = 9.385 + 0.021x r ² = 0.514 8
16
1 2 3 4 5 6 7 8 9 101112131415161718 12
40
Trial
1 2 3 4 5 6 7 8 9 101112131415161718 20
Driving frequency: 1.000
Driving frequency: 1.000
11
19
10
18
9
Trial
y = 10.409 + 0.02x r ² = 0.275
17
y = 17.768 + 0.069x r ² = 0.863
16
8 1 2 3 4 5 6 7 8 9 101112131415161718
Trial
1 2 3 4 5 6 7 8 9 101112131415161718
Trial
Figure 4. Mean distance of segments across trials in each condition (line length by driving frequency). The dotted line is the line length for the conditions. Included in each panel is the outcome of linear regression. Vertical bars indicate the standard error of the means. Note: the ordinal scales vary according to segment length and driving frequency, and the 0.0265 Hz trials have twice the duration.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
41
should be half the duration of the driving frequency. The movements during the 16 cm conditions were around the ideal performance (Table 1). However, in the 8cm condition, the duration was significantly smaller then expected. This indicates that the main effect was to decrease duration as line length decreased. In relative terms, the reduction in duration was minimal and ranged from 6% to 11% depending on condition. The lack of main effects of trial, suggest that the expected practice effects may not of happened. However, in Figure 5 there appears to be practice effects in the high frequency conditions, for the first few trials. Note that in the 8 cm high frequency condition, these practice effects reduced the duration even more then expected, which indicates the main effects were due to trial effects for this condition. The effects observed may be significant, but not enough to effect the interaction between driving frequency and trial. In addition, Figure 5 supports the findings from the Table 1 that movement during the 16 cm conditions, were around ideal performance, with some practice required in the fast and medium conditions to obtain this performance. Combined with the ANOVA, the effects of line length on duration were to reduce the shorter movements below optimum. Speed. All main effects were significant: line length F(1, 17)=1668.02, p<.001, MSE=11.25, driving frequency F(2, 34)=2304.83, p<.001, MSE=61.87 and trial F(17, 289)=8.26, p<.001 MSE=2.26. In addition, all interactions were significant line length by driving frequency F(2,34)=1254.43, p<.001 MSE=5.91, line length by trial F(17,289)=3.17, p<.001, MSE=1.34, driving frequency by trial F(34,578)=7.2, p<.001, MSE=1.98, as well as the complete model F(34,587)=2.25, p<.001, MSE=1.33. These follow expectations that movements were executed at different speeds in each condition, and indicate the possible effects of practice. Combined with Table 1, driving frequency had a positive relationship with speed and the overshoots, as was expected. The effects of distance were partially expected and increased the speed of movements. However, the shorter movements had an
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
LIne Length: 8 cm
42
Line Length: 16 cm 9
Driving frequency: 0.0625
Driving frequency: 0.0625
8
y = 8.076 + 0.003x r ² = 0.016
8 7
y = 7.376 - 0.012x r ² = 0.186 7 1 2 3 4 5 6 7 8 9 101112131415161718
1 2 3 4 5 6 7 8 9 101112131415161718 2.1
Trial
Trial 2.3
Mean Duration (s)
Driving frequency: 0.250 2.2
2.0
Driving frequency: 0.250 y = 2.067 + 0.006x r ² = 0.016
2.1
1.9 2.0
1.8
y = 1.858 + 0.002x r ² = 0.173
1.7
1.9 1.8
1 2 3 4 5 6 7 8 9 101112131415161718
Trial
1 2 3 4 5 6 7 8 9 101112131415161718 0.65
Driving frequency: 1.000
Trial Driving frequency: 1.000
0.60
0.50
y = 0.53 + 0.002x r ² = 0.471
0.55
0.45
y = 0.481 + 0.001x r ² = 0.54
0.50
0.45
1 2 3 4 5 6 7 8 9 101112131415161718
Trial
1 2 3 4 5 6 7 8 9 101112131415161718
Trial
Figure 5. Mean duration of segments across trials in each condition (line length by driving frequency). The dotted line is the ideal duration for the conditions. Included in each panel is the outcome of linear regression. Vertical bars indicate the standard error of the means. Note: the ordinal scales vary according to segment length and driving frequency, and the 0.0265 Hz trials have twice the duration.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
LIne Length: 8 cm 1.6
2.5
Driving frequency: 0.0625
1.5
Line Length: 16 cm Driving frequency: 0.0625
y = 1.23 - 0.001x 2.4 r ² = 0.014 2.3
1.4
y = 2.15 + 0.002x r ² = 0.051
1.3
2.2
1.2 1.1
2.1
1.0
2.0 1 2 3 4 5 6 7 8 9 101112131415161718
1 2 3 4 5 6 7 8 9 101112131415161718
Mean Speed (cm/s)
6
Trial
Trial 10
Driving frequency: 0.250
Driving frequency: 0.250
5
9
y = 8.73 + 0.029x r ² = 0.404
y = 5.10 + 0.004x r ² = 0.076 4
26
8
1 2 3 4 5 6 7 8 9 101112131415161718
1 2 3 4 5 6 7 8 9 101112131415161718
Trial
Trial 40
Driving frequency: 1.000
24
Driving frequency: 1.000
38
22
36
20 y = 21.83 + 0.095x r ² = 0.477
18
y = 34.15 + 0.255x r ² = 0.716
34 32
16
30
14 1 2 3 4 5 6 7 8 9 101112131415161718
Trial
1 2 3 4 5 6 7 8 9 101112131415161718
Trial
Figure 6. Mean speed of segments across trials in each condition (line length by driving frequency). The dotted line is the ideal speed for the conditions. Included in each panel is the outcome of linear regression. Vertical bars indicate the standard error of the means. Note: the ordinal scales vary according to segment length and driving frequency, and the 0.0265 Hz trials have twice the duration.
43
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
44
overall proportional increase in speed, which would likely caused by the shorter durations. For example the medium frequency conditions for the 16 cm compared to the 8 cm had a ration of 1:1.14. If these were in proportion it would be closer to 1:0.5. The effects of practice were not as expected, in the conditions in which it occurred, speed appeared to increase with trial away from ideal performance (Figure 6). The complete interaction appeared to be that driving frequency and line length had a negative relationship with practice effects on speed. Together including the complete model, this indicates that it was easer to control speed, as movements were aimed at lower durations over longer distances. Four of the additional five variables indicating stationary movement are presented in Table 2. The results indicate: dwell periods occurred often between segments and had a negative relationship with line length and driving frequency. Dwell time had a negative relationship with driving frequency. In absolute terms dwell time indicates the duration of the periods were rarely in excess of 0.25 s, which is relatively minimal as expected. The trials with stops periods indicate these periods did occur in a large percentage of trials, and this percentage increased as driving frequency decreased. Stops periods did not occur often in the high frequency condition as would be expected. The average number of stop periods increased as driving frequency decreased, and distance only increased the number during the low driving frequency. In the low frequency condition stop periods occurred between one to two times a segment, which was considered often. Stop time had a negative relationship with driving frequency, and distance increased the duration during the low driving frequency and likely during high. In absolute terms the duration of these was unexpected, particularly in the low driving frequency, since the duration was large enough to affect the speed displayed in Figure 6, indicating the movements within the segment were actually faster. The number of regressions (not reported in Table 2) indicates at least one regression occurred in 3.29% of the trials, with 1.13% occurring in the 8 cm low driving frequency condition alone. Together,
86.93% (4.21)
95.68% (2.40)
0.25 Hz
0.0625 Hz
70.60% (4.89)
90.74% (2.73)
0.25 Hz
0.0625 Hz
0.222 (0.03)
0.110 (0.01)
0.068 (0.003)
0.258 (0.02)
0.136 (0.01)
0.073 (0.003)
Dell Time (s)
60.49% (6.68)
32.10% (4.15)
20.99% (5.01)
73.56% (6.04)
31.17% (4.54)
24.38% (3.45)
M (S.E.)
Trials with Stop Periods
1.29 (0.27)
0.32 (0.03)
0.07 (0.01)
1.89 (0.43)
0.31 (0.03)
0.07 (0.01)
Stop Periods (n)
0.43 (0.09)
0.13 (0.02)
0.08 (0.01)
0.68 (0.15)
0.12 (0.01)
0.06 (0.01)
Stop Time (s)
having an average of 1.89 stop periods per segment for an average duration of 0.68 seconds.
Notes: An example for interpreting ‘stops’ is: stops periods occurred in 73% of the trials for one condition with each of these trials
30.34% (5.06)
1 Hz
16 cm
52.67% (5.55)
1 Hz
8 cm
Conditions
Dwell Periods
(8cm, 16cm) by Driving Frequency (high 1Hz, medium 0.25Hz and low 0.065Hz).
Summary of Stationary Movement Estimates (Standard Error) for participants’ performance. The six conditions are Line Length
Table 2
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS. 45
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
46
these results indicated that sustaining the movement throughout the trial became harder, and the movements became more discrete, as line lengths and driving frequency decreased. Overall, participants were able to follow the expectations of the conditions with four exceptions. The expected performance was based on the previous research indicating deviations from ideal performance that included: overshoots, higher speed and minimal duration of stationary movement (Atkeson & Hollerbach, 1985; Mathur & Hughes, 2009; Morgan, et al., 1994). The stationary movement occurred often (dwell periods, stops) and the duration of dwell time was minimal. The stop time length for some conditions, particularly the low frequency, was unexpected and a possible exception because it indicates that the mean speed or velocity of non-stationary movement was higher then the speed measured and stationary movement was not ideal. However, these stationary movements are also a part of the intermittent profile and accounted for in the normalized jerk-cost measure, of intermittency (Hogan & Sternad, 2009). The first exception then was the significantly shorter duration of the segments in the short line conditions, which likely increased the speed of these segments due to their relationship. This indicates that as the ideal distance decreases, the ability to meet the ideal duration is reduced. In terms of driving frequency, this means that the movement cycles occurred at a higher frequency then ideal. The other three exceptions were the effects of practice that influence the movements’ distance, duration and speed away from ideal performance in some conditions. These findings indicate that during these conditions, participant’s ability to control these aspects of their movements reduced. These four exceptions likely had a marginal effect on intermittency, and this will be discussed. It was easer to control speed, as movements were aimed for longer durations over longer distances. Thus in summary, these results indicate that people could do the task, and as driving frequency and line length decreased it became harder to maintain a continuous movement, whereas control of average speed was easier for lower frequencies and longer line
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
47
lengths. Movement Intermittency The profiles of Kryten’s performance in each condition (Figure 3) provide a visual example of intermittency under the various conditions. The profiles indicate the increase in velocity fluctuations as driving frequency decreased and there appears to be little effect of distance on intermittency. The degree of intermittency does not appear to have changed between the profiles of the first trial and the profiles of the last trials. This suggests that repeated opportunity to practice did not affect Kryten’s velocity traces. The analysis of the quantitative measure of intermittency, acceleration zero-crossings and normalized jerk-cost, extent to which a movement is executed in a smooth and constant manner. Acceleration zero-crossings. The complete factorial repeated measures 2 x 3 x 18 (Line Length [8 cm, 16 cm] x Driving Frequency [0.0625 Hz, 0.25 Hz, 1 Hz] x Trial [1 to 18]) ANOVA was conducted with average number of acceleration zero-crossing per segment as the dependent variable. The hypothesis that distance would have no effect on intermittency was supported with a no effect of line length F(1, 17)=2.76, p=.115, MSE=199.02. Two significant main effects of driving frequency F(2, 34)=1041.84, p<.001, MSE=1437.78 and of trial F(17, 289)=1.88, p=.02 MSE=51.12 were found. These two main effects provide initial support for the hypothesised effects of driving frequency and practice on intermittency. Figure 7 adds to this by depicting the lack of change due to line length and a possible negative curvilinear relationship with driving frequency. There was one statistically significant interaction. This was between driving frequency and trial on acceleration zero-crossings F(34,578)=1.5, p=.037, MSE=50.81. This interaction is depicted in Figure 8 and indicates how the relationship between acceleration zero-crossings and driving frequency was affected by trial. This provides initial support for the hypothesized relationship between for driving frequency and practice. There was no
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
48
120
Mean no. of Acceleration Zero-Crossings
8 cm line 16 cm line
100
80
60
40
20
0 0.0
0.2
0.4
0.6
0.8
1.0
Driving Frequency (Hz)
Figure 7. Mean acceleration zero-crossings as a function of driving frequency for both line lengths. Vertical bars indicate the standard error of the means.
interaction between line length and trial F(17,289)=1.42, p=.124, MSE=47.32 which indicates the hypothesized relationship between distance and practice may not supported. No other interactions were significant. The polynomial contrast effects were conducted between the significant relationships, to test if the trends followed the hypotheses. The hypothesis; that there would be a curvilinear relationship between driving frequency and intermittency, was supported with the main effect having significant linear F(1, 17)=1065.74, p<.001, MSE=2804.6 and quadratic F(1,17)=96.98, p<.001, MSE=70.96 contrast effects. Incorporating this with visual analysis of Figure 7 the significant curvilinear manner was negative as hypothesized.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
Line Length: 16 cm
Line Length: 8 cm 110
110
Driving frequency: 0.0625
Driving frequency: 0.0625
100
100
90
90
y = 98.99 - 0.517x r ² = 0.570
y = 95.61 - 0.124x r ² = 0.060 80
Mean No. of Acceleration Zero Crossings
80 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
Trial 32
26
Trial 32
Driving frequency: 0.250
30 28
Driving frequency: 0.250
30 28
y = 22.87 - 0.008x r ² = 0.008
26
24
24
22
22
20
20
18
y = 27.76 -0.132x r ² = 0.411
18 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
Trial
Trial 3.0
3.0
Driving frequency: 1.000
Driving frequency: 1.000
2.5
2.5
2.0
2.0
1.5
49
y = 2.28 + 0.022x r ² = 0.501
1.5
y = 2.20 - 0.035x r ² = 0.606
1.0
1.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
Trial
Trial
Figure 8. Estimates of acceleration zero-crossings across trials (line lengths by driving frequencies). Included in each panel is the outcome of linear regression. Vertical bars indicate the standard error of the means. Note: the ordinal scale varies by driving frequency, and the 0.0265 Hz trials’ have twice the duration.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
50
Did practice effects occur with acceleration zero-crossing? The contrast effects for the main effect of trial on acceleration zero-crossings had significant contrast effects, of both linear F(1, 17)=4.95, p=.04, MSE=199.57 and Order 167. The linear contrast indicated that the decrease (Figure 8) in acceleration zero-crossings was constant across trials, regardless of effect of condition. The order 16 contrast detected significant fluctuations across trials. These results indicate the occurrence of practice effects. The hypothesised interaction between driving frequency and practice on intermittency was neither supported nor denied support, by the contrast effects. These contrast effects were: linear driving frequency, order 7 trial F(1, 17)=6.05, p=.025, MSE=62.27, quadratic driving frequency, order 6 trial F(1, 17)=8.63, p=.009, MSE=5.65 and quadratic driving frequency, order 7 trial F(1, 17)=4.66, p=.046, MSE=6.67. These contrasts indicate that the curvilinear relationship between driving frequency and acceleration zero-crossings was found across trials, and trials affected this relationship in a fluctuating manner. What this indicated was that practice effects did not occur across all of the driving frequency conditions but occur in some of the conditions. Visual analysis8 was conducted on Figure 8 with a focus on trend analysis (See Doeringer & Hogan, 1998; Khan, et al., 1998; Pratt, et al., 1994 for previous examples) to determine in what conditions the effects of practice occurred. Visual analysis was also chosen as the analysis indicated disagreement with the regression analysis and the hypothesized effects of duration and practice on intermittency. At each driving frequency did the longer line conditions have more effects of practice? Practice effects occurred in the high and low frequency conditions. In the low conditions, there were practice effects for the longer line but not for the shorter line. In the high conditions, the reduction due to practice was greater for the 16 cm line. In the medium conditions, although there was no decrease in 7
For simplicity, the statistics for significant order effects (α = 0.05), above cubic are not reported, and are available upon request. 8 For more information (Cooper, Heron, & Heward, 2007) provide a well established guide on how to conduct a visual analysis of trend and change in baseline.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
51
trend of the mean, there appears to be a week decrease in the trend of the standard error. This medium condition also shows that even through the means were statistically similar, between the line lengths, the first few trials of the longer line had higher number of acceleration zerocrossings. These observations support the hypothesis between distance and practice on intermittency. When practice effects occurred, did the hypothesised relationship between driving frequency and intermittency occur? The low frequency longer condition did have practice effects across all trials. However, the hypothesis was unsupported because there were no practice effects in the medium condition. The high conditions had noticeable effects only across the first few trials, three (8 cm) and four (16 cm) and then levelled out. In absolute terms, these high conditions had a average count between 1.5 to 2 (1.5 acceleration zero-crossings indicates that a large proportion had at least two crossings) after levelling out and did not generally get close to one, suggesting that floor effects did not happen. This does not support the expectations of floor effects in these conditions. There may have been a couple of individuals’ closes to one most of the time, but this unclear based on analysis of Figure 8. An alternative way to gauge the effects of practice is to estimate the reduction as a percentage of the starting trials. An advantage when using this method is that it is not affected by scale. A comparison between the first and last trial (Table 3), gives a rough quantitative estimate of the effects of practice and alters the interpretation of the results, perceived in Figure 8. The hypothesized interaction between driving frequency and practice on intermittency was neither supported nor denied support, by the percentages in Table 3. This was because low frequency movements had proportionally less improvement then high frequency movements. However, this is limited to 18 trials and the practice effects may continue with more trials, particularly in the low condition because it has yet to level out.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
52
Table 3 The Percentage that Changed in Acceleration Zero-Crossings from the first to the last trials for the six conditions and also a Summary of the Visual Analysis. Line Length Driving Frequency
Short (8 cm)
Long (16 cm)
23.09%*
25.99%*
Medium (0.25 Hz)
1.67%
6.59%
Low (0.0625 Hz)
4.71%
10.95%*
High (1 Hz)
*Conditions in which practice effects occurred based on the visual analysis.
Normalized jerk-cost. To initiate the investigation of the hypotheses with average normalized jerk-cost per segment, the complete factorial repeated measures 2 x 3 x 18 (Line Length [8 cm, 16 cm] x Driving Frequency [0.0625 Hz, 0.25 Hz, 1 Hz] x Trial [1 to 18]) ANOVA was also conducted. Two significant main effects, of driving frequency F(2, 34)=229.46, p<.001, MSE=353258055.2 and trial F(17, 289)=3.28, p<.001 MSE=15306073.19 were found. No effect of line length emerged F(1, 17)=4.39, p=.051, MSE=39327021479 was found. As for acceleration zero-crossings, these results support the hypothesised relationship of distance and intermittency, and the partial support for both the negative curvilinear relationship for duration and intermittency (Figure 9) and the practice effects. Three interactions were statistically significant. The first was an interaction (Figure 9) between line length and driving frequency F(2,34)=4.28, p=.022, MSE=73380213.39. This result was not hypothesized, but suggest that short low frequency movements had higher
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
53
25000
8 cm line 16 cm line
Mean Normalized Jerk-Cost
20000
15000
10000
5000
0 0.0
0.2
0.4
0.6
0.8
1.0
Driving Frequency (Hz)
Figure 9. Mean acceleration zero-crossings as a function of driving frequency for both line lengths. Vertical bars indicate the standard error of the means.
jerk-cost then long low frequency movements. The second was between driving frequency and trial F(34,578)=3.1, p<.001, MSE=15139706.82. This provides initial support for the hypothesized relationship between driving frequency and intermittency. The three-way interaction (Figure 10) was also significant F(34,578)=1.5, p=.036, MSE=19010153.71 which was not hypothesized but indicates all three variables affected intermittency. The no significant interactions between and line length and trial F(17,289)=1.49, p=.096, MSE=19087804.51 which indicates the hypothesized relationship between distance and practice may not be supported. Polynomial contrast analysis shows a curvilinear relationship between driving frequency and intermittency was supported with significant linear F(1, 17)=229.88, p<.001,
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
Line Length: 8 cm
Line Length: 16 cm 4e+4
4e+4
Driving frequency: 0.0625
Driving frequency: 0.0625 3e+4
3e+4
2e+4
2e+4
1e+4
y = 433464.61 - 5024.051x r ² = 0.172
1e+4
y = 243608.61 - 3384.852x r ² = 0.277
0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
Trial
Mean Normalised Jerk-Cost
54
1400
Trial 1400
Driving frequency: 0.250
1200 1000
1000
800
800
600
600
400 200
400
y = 514.52 - 4.807x r ² = 0.173
200
0
Driving frequency: 0.250
1200
y = 401.50 - 11.905x r ² = 0.601
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
Trial 50
Trial 50
Driving frequency: 1.000
Driving frequency: 1.000
40
40
30
30
20
20
10
y = 0.56 - 0.009x r ² = 0.459
10
0
y = 0.48 - 0.014x r ² = 0.472
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
Trial
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
Trial
Figure 10. Estimates of normalized jerk-cost across trials (line lengths by driving frequencies). Included in each panel is the outcome of linear regression. Vertical bars indicate the standard error of the means. Note: the ordinal scale varies by driving frequency and driving frequency, and the 0.0265 Hz trials have twice the duration.
MSE=688682525.6 and quadratic F(1, 17)=213.26, p<.001, MSE=17833584.84 contrast effects. This curvilinear relationship was also negative as expected, but it was also affected
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
55
by line length (Figure 9). This unexpected relationship with line length only occurred during the low driving frequencies with 16cm (M=18582.51, SE=1441.22) movements being smoother then the 8 cm (M=21010.28, SE=1404.27) movements. Did practice effects occur with normalized jerk-cost? Yes the main effect of trial on normalized jerk-cost had significant contrast effects, of both linear effects F(1, 17)=13.41, p=.002, MSE=28684331.04, Order 16, and Order 8. As with acceleration zero-crossings, these indicate that normalised jerk-cost decreased across trials, regardless of the effects of condition (Figure 10). In contrast to the acceleration zero-crossings, the hypothesized interaction between driving frequency and practice on intermittency was neither supported nor denied by contrast effects. These contrast effects were linear driving frequency, linear trial F(1, 17)=12.47, p=.003, MSE=50743464.85, quadratic driving frequency, linear trial F(1, 17)=7.508, p=.014, MSE=1647458.17 and significant contrast effects of Order 6, 7 and 16 trial for both linear and quadratic driving frequency effects. The contrasts indicated that trial had a negative linear relationship with the curvilinear relationship of driving frequency. The complete interaction with line length, driving frequency and trial also had significant contrast effects of linear driving frequency with cubic F(1, 17)=4.59, p=.047, MSE=30381622.93 and order 12 trial, and quadratic driving frequency with trial order 12 trial. This indicated that distance impacted the interaction between driving frequency and trial, by making the effects of trial more variable. Visual analysis of the trend across trials of Figure 10 follows. The practice effects hypothesis between distance and intermittency was supported and based on the following observations. Practice effects occurred in the medium and high frequency conditions. In the medium conditions, there were practice effects for the longer line but not for the shorter line. In the high conditions, the results were similar to acceleration zero-crossings, although the greater decrease in the 16 cm condition does not appear as to be much for normalized jerk-
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
56
cost and there is no evidence of floor effects at all. In the low conditions, although there were no practice effects, there was a sudden decrease after the 15th trial for the longer line. The sudden decrease in this condition maybe related to variability masking a decrease in trend, however this was unclear and thus the condition was considered to not have practice effects. A closer inspection of the participants’ data suggested that the drop was a related to a number of individuals and not outliers. In addition, the visual analysis elucidates the interaction between line length and driving frequency for low conditions indicating trail effects were responsible, due to some trials in the short condition having high levels of intermittency and the sudden decrease in the long condition. The practice effects hypothesis, between driving frequency and intermittency was also unsupported for normalized jerk-cost. This hypothesis was unsupported, in contradiction to the contrast effects, because the effect of trial in the low conditions was not considered to have practice effects. As for acceleration zero-crossings, the changes between the first trial and the last were compared (Table 4), in terms of percentage and doing so alters the pattern found. Table 4 The Percentage that Changed in Normalized Jerk-Cost from the first to the last trials for the six conditions and in addition a Summary of the Visual Analysis. Line Length Driving Frequency
Short (8 cm)
Long (16 cm)
High (1 Hz)
19.85%*
33.04%*
Medium (0.25 Hz)
14.27%
32.55%*
Low (0.0625 Hz)
8.92%
28.74%
* Conditions in which practice effects occurred based on the visual analysis.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
57
Similar to acceleration zero-crossings, the visual analysis was supported by the percentage change in means, for distance and practice effects. However, for normalized jerk-cost the percentage change indicates a greater decrease then visual analysis, for the both high conditions. Velocity. The two measures of intermittency were then modelled with respect to the actual average velocity of the segments in a trial. Acceleration zero-crossings (Figure 11) Normalized jerk-cost (Figure 12). The exponential functions was chosen because they appear
Mean No. of Acceleration Zero Crossings
200
Right >> Left
Left >> Right
150
100
y = 3.67 + 148.43 * exp(-0.30 * x) 2 r = 0.865
y = 3.77 + 130.89 * exp(-0.27 * x) 2 r = 0.874
50
0 -50
-40
-30
-20
-10
0
10
20
30
40
50
Mean Velocity (cm/s)
Figure 11. The relationship between the mean acceleration zero-crossing and mean velocity for the segments is presented. Each data point represents either the leftward or rightward segments of the cycles, during a single individual’s trial. In addition, the results of nonlinear (decreasing exponential) functions are displayed. The dashed lines are the ideal velocities for each condition.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
58
1e+5
Left >> Right
Mean Normalized Jerk-Cost
Right >> Left 8e+4
6000 5000 4000
6e+4
3000 2000
4e+4
1000 0 -30
2e+4
y = - 294.77 + 38675.47 * exp (0.4543 * x). r2 = 0.671
-20
-10
0
10
20
30
y = - 244.59 + 48801.66 * exp (0.54 * x). r2 = 0.644
0 -50
-40
-30
-20
-10
0
10
20
30
40
50
Mean Velocity (cm/s)
Figure 12. The relationship between the mean normalized jerk-cost and mean velocity for the segments is presented. Each data point represents either the leftward or rightward segments of the cycles, during a single individual’s trial. In addition the results of nonlinear (exponential) functions are displayed. The dashed lines are the ideal velocities for each condition. The insert depicts a scaled version of relationship between normalized jerk-cost and mean velocity. Note: one data point (x = 0.32, y = 152436.02) was not displayed.
to describe the data and although not reported here other models were fitted, but the other models generally had r-squared values that were either: lower (polynomial inverse second/third order) or the increase due to an additional parameter (exponential with four parameters) was marginal. Both figures indicates the relationship between intermittency and velocity can be represented by an inverse exponential function, as was expected. Combining both measures (Table 3 and 4) indicates that according to visual analysis practice effects occurred on intermittency in all of the long line conditions and just the short
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
59
high frequency condition. The practice effects and the percentage change in the longer line compared to the short line and the difference between the high conditions supports the distance and practice hypothesis. Line length, appears to determine if practice effects occurred in 18 trials of the lower frequency conditions and the degree or amount of reduction due to practice in the high driving frequency conditions. These findings also indicate that driving frequency appears to determine the occurrence of practice effects in the high short condition when compared to the short medium condition. This may also be the case for the longer line condition since they have similar velocity profiles and levels of intermittency. Together duration and distance appear to influence the high frequency movements. The percentage change between the first and the last trail indicates that when practice effects occur the reduction is between 10-34% depending on driving frequency and line length. In summary, a majority of the hypothesis were supported. The hypothesis of no distance effects on intermittency was supported by both measures of intermittency. The hypothesized negative curvilinear relationship between driving frequency and intermittency, or velocity and intermittency were both supported. Practice effects were found for both measures. The visual analysis indicated the hypothesized effects of line length on the ability to practice were also found for both measures. The effects of driving frequency and practice on intermittency for either measure were not supported. Both measures had at least one condition without practice effects, either medium (acceleration zero-crossing) or low (normalized jerk-cost) frequency conditions. In addition, expected floor effects may have been found for some participants, but only for acceleration zero-crossings. Two finding for normalized jerk-cost were not hypothesized, these were the complete model and the effects of line length on curvilinear relationship between driving frequency and intermittency. Overall, driving frequency appears to determine the level of intermittency and distance and driving frequency determine practice effects.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
60
Discussion In the present study, participants were blindfolded and asked to move at “as smooth and constant a velocity” under three timing and two distance constraints, with the availability of haptic feedback. The aim was to determine if people could consciously improve the smoothness of their movements in the presence of this haptic information. This aim enables the investigation of the contribution of cognitive or neurobiomechanical sources that creates movement intermittency. The aim was split into two goals: to determine if participants can move (or can learn to move) continuously at a constant velocity under the different timing and distance constraints, and second to determine the effects of the factors duration, distance and practice on intermittency. In addition, the study design and results could provide information for creating methods, which could be used to improve the performance of similar tasks. Research indicated (1) that speed was a factor in influencing intermittency and that as speed decreases intermittency increases (Celik, et al., 2009), (2) that practice smoothed nonvisually guided movements during tasks requiring fast accurate movements (Khan & Franks, 2000) or timing constrained accurate movements (Proteau & Cournoyer, 1990); and (3) that haptic information could be used to detect online velocity (Dépeault, et al., 2008). The present research indicated that people could generally perform the task with modest expected systematic error. The primary factor in determining the degree of intermittency was driving frequency, and driving frequency interacted with line length on the ability to improve the smoothness of movements with practice. Five of the hypotheses were supported and the interaction between driving frequency and practice effects on intermittency was partially supported. Evidence for the hypothesis that participants could move as required under the different conditions of the study came from their performance matching the expected performance with marginal unexpected
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
61
performance results. The expected systematic distortions from ideal performance were found and these were overshoots, higher speed, and minimal duration of stationary movement at the end of the line (Atkeson & Hollerbach, 1985; Mathur & Hughes, 2009; Morgan, et al., 1994). There were four unexpected distortions: for example the shorter line had shorter movement duration (which likely increased the speed of these movements), and counter intuitively practice effects could influence the movements’ distance, duration and speed away from ideal performance in some conditions. These findings indicate that during these conditions, participants’ ability to control these aspects of their movements reduced marginally, but they could still perform the task. The participants found shorter distances and lower driving frequencies more difficult to execute based on the relationship between both the unexpected results and the stationary movement with driving frequency and line length. As instructed, participants did not, avoid the slower movements nor did they have relatively extensive dwell time (van der Wel, et al., 2010). Taking in to account the small increases in difficulty the hypothesis was supported. The findings of no statistical main effect of line length on intermittency supported the hypothesis that distance under fixed driving frequencies would have no effect on intermittency for both measures. This was observed in velocity profiles with the main difference between the line lengths being an increase in velocity. The hypothesis that driving frequency and intermittency would follow a negative curvilinear relationship was supported. Evidence for this hypothesis, was from finding a significant negative curvilinear trend across driving frequencies for both measures. The results also indicated that line length had a significant effect on this relationship for the normalized jerk-cost measure in the low frequency conditions. The additional investigation supported this hypothesis by finding an inverse exponential function fitting the relationship between velocity and intermittency.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
62
Significant negative linear contrast effects supports the hypothesis for the occurrence of practice effects across trials for both measures of intermittency: normalized jerk-cost and acceleration zero-crossings. The results suggested that when practice effects occurred, distance and practice had the hypothesized relationship. The hypothesis was that as line length increased the ability to improve the smoothness of the longer movements would increase, compared to the shorter movements. The analysis found that practice effects either occurred only for the 16 cm line length conditions in the lower conditions or had greater reductions in intermittency between the high conditions. The hypothesised effects of duration on intermittency were partially supported. The hypothesis was that movements would become smoother in the medium and low driving frequency conditions. This hypothesis was combined with the expectation that if there were practice effects in the high frequency conditions, these would be minimal due to floor effects. The hypothesis was supported if results of both measures of intermittency are combined. The expectation on the high frequency conditions, however, was unsupported because these conditions had practice effects which were unaffected by floor effects. (The possible acceleration zero-crossing floor effects for some participants will be further explored in the discussion on movement intermittency.) The results indicated that there was a reduction of around 10-30% depending on both driving frequency and line length. However, the hypothesis was not supported if the results are separated out by the different intermittency measures. This was because both measures had a condition without practice effects, the condition for acceleration zero-crossings was the medium condition and for normalized jerkcost was the low condition. Thus driving frequency appeared to determine the practice effects on the different components of intermittency for the lower frequency movements. In summary, the results suggest duration determined the degree of intermittency; distance determined occurrence of practice effects in the 18 trials for lower frequency movements; and
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
63
duration determined the components that were improved in the medium or low frequency conditions whilst both distance and duration determined the practice effects for high frequency conditions. Task Performance This section focuses on an analysis of the results with the aim to provide possible explanations and implications for: some of the expected systematic deviations from ideal performance and the unexpected findings. This section was explored prior to intermittency for two reasons: the analysis suggests that some of the conditions are controlled differently than what was assumed during the hypotheses, and exploring the controlling variables of the movement error in general provides a foundation for exploring the controlling variables for the main movement error measured, the intermittency. It was assumed during the hypothesis that driving frequency was the primary variable for movement error, particularly intermittency, due to driving frequencies’ relationship with velocity. This section suggests that the controlling variable for error in overall performance was generally duration, which has a more direct relationship with driving frequency. The expected systematic changes in movement were overshoots: unexpected were duration, traveling speed, stationary movement and practice effects. The overshoots indicate a difficulty for people to determine when to reverse direction, which could result from planning the movements (due to being blindfolded), auditory feedback, online haptic feedback, or a combination. Of these factors, a planning issue during high frequency movements would result from being blindfolded; this could create prior to moving uncertainty of where the end of the line was before executing the ballistic movements. To resolve this uncertainty, the first couple of movements of a high frequency trial were slower (and more intermittent) to determine the length of the line (Figure 13). This could suggest the control of these high frequent movements may not be controlled by
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS. Segment Length: 8 cm
Segment Length: 16 cm
70
70 High freq. v1 High freq. v2
60
Velocity (cm/s)
64
50
60 50
40
40
30
30
20
20
10
10
0
0 4
6
8
10
12
14
Finger Position (cm)
16
2
4
6
8 10 12 14 16 18 20 22 24 26 28 Finger Position (cm)
Figure 13. Velocity profiles of two participant’s movements traveling left to right. The Driving Frequencies for the movements (from left to right then back) were the high, 1 Hz. The high freq. v2 exemplify a first segment of a trial. The high freq. v1 exemplify a segment after the first few segments of a trial. Dwell time was removed from each trace. The grey square represents the raised lines. duration and controlled differently to the other lower frequency movements. However, this perspective does not rule out duration due to the possibility that each ballistic submovement was executed based on a specific duration. The overshoots during the lower frequency occurred even though there was plenty of time to stop during execution of the movements signifying that the overshoots were not an error in planning. This observation supports the idea that low frequency movements are controlled differently to high frequency movements. The online haptic feedback was a likely factor for the participants overshoots in all of the driving frequencies. As it appeared they could not distinguish the end of the line until there was no more line, at which point their fingers had gone past the end and would then decide to stop. Similar overshoot distances between the same driving frequencies seem to support this theory, based on the idea that traveling at relatively similar speeds results in similar stopping distances or overshoots (Table
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
65
1, Figure 4 and 6). Accordingly, haptic feedback could be influencing movements to create overshoots at all three driving frequencies. The overshoots in some trials, particularly the low conditions, had increased or maintained a relatively similar velocity values for short periods after the end of the line (Figure 3 and Figure 17 in the section on acceleration zero-crossing and normalized jerkcost), which neither the planning while blindfolded or haptic feedback perspectives can explain. This behaviour could result from participants prioritizing the timing of the beats for each movement over distance, since they changed direction after the required beats were reached instead of changing direction at the end of the line. In these situations they could well have used the beat as auditory feedback, rather then the haptic feedback, to guide changes in direction (the distance travelled by the movements). Once timing was met, participants would move their finger quickly back to the line (Figure 17), suggesting they then used haptic feedback to start the return movement. The prioritizing of timing found for the long high frequency conditions with practice effect suggest that these movements became more ballistic to meet the timing requirements as the participants learned the length of the line. Thus duration (set by the auditory feedback of the metronome) was the likely controlling variable in these situations. Together these ideas of planning, haptic feedback and auditory feedback point to three possible explanations for the overshoots and that duration and speed may control the higher frequency movements. The unexpected finding of shorter duration of movements in the short line conditions could be due to two factors, neurobiomechanical constraints or experience and the relationship these short line conditions have with auditory and haptic feedback. The shorter duration results suggest that participants had difficulty with these conditions regardless of driving frequency (Table 1). Likely explanations for these results were participants’ lack of experience with the shorter distance or the motor system (neurobiomechanical constraints)
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
66
has not evolved effectively for these movements. The research by Shadmehr and MussaIvaldi (1994; von Hofsten, 1991; von Hofsten & Rönnqvist, 1993) provide support for the idea that the majority of or possibly all of this difficulty was experienced based, and these short movements are simply a form of novel environment (with an internal locus of control over the constraints rather then external). In the medium and low frequency short distance conditions, participants appeared to be learning the duration requirements but had yet to achieve them in the 18 trials (Figure 5). This suggests that participants could not reduce the intermittency because they had yet to finish learning the task. This was observed by a decrease in the variability of intermittency in Figure 10, and more trials would be necessary to further support a general increase in mean to ideal. Where as in the short high frequency condition, participants avoided ideal duration by decreasing it. These learning behaviours, which will be discussed more under practice effects, provide a reason for the overall short conditions’ having significantly lower duration. Together the learning and avoidance behaviour suggests that experience or neurobiomechanical constraints are on a continuum, with one being in the opposite direction of the other as driving frequency or distance increases. Either experience, constraints or both can be influencing participants’ ability to perform the task in the short distance conditions. The unexpected shorter duration provides inferences for the haptic and auditory feedback loops. If the metronome were guiding the movements during the short distance more than haptic feedback, then duration would have been, as expected, similar to the longer line conditions. Since this did not occur, these short distance movements are being influenced by the distance information provided by the raised line (haptic feedback) more than the auditory feedback from the metronome. The travelling speed of movement between the ends of the line were higher than ideal, proportionally higher for the short distance, and had unexpected practice effects. These three
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
67
findings on speed in the six conditions can be explained by the changes in both the duration and distance of the movements, as might be expected based on the co-varying relationship between these variables during a constant velocity task, as also suggested by van der Wel, et al. (2010). The stationary movement that changed with frequency and duration across the conditions indicate a change in the nature of the movements from continuous high frequent cyclic movements to more discrete (stop period, move period, stop period) movements, as duration increases and distance decreases (Table 2). This change in movement execution was likely due to a change in method of control. For example, the dwell time results are compatible with previous research that dwell time increases as speed decreases (Morgan, et al., 1994). Dwell time could indicate the time required to prepare for the upcoming movement. Hence, more preparation would be needed for the movements of each segment within lower frequency trials. Alternatively, dwell time could indicate a preference to be ‘stationary’ for as long as possible before moving more rapidly (van der Wel, et al., 2010). This alternative perspective also provides a possible explanation for the stop time, which was larger then dwell time in the lower frequency conditions. These perspectives provide two possible ideas for future research. The practice effects, from the task performance data, appeared to depend on the conditions (Figures 4, 5 and 6). These effects indicated that duration was the controlling variable as it was the only variable of the three to have ideal performance or improved performance, in all but the short high frequency condition. Weaker evidence also comes from duration being the only variable to have a negative relationship with the other two variables, in some of the conditions. This evidence was weaker because it was only an indicator that duration could be more independent across trials then distance and velocity, which change together.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
68
The practice effects do provide marginal evidence for velocity being the controlling variable, in the short high frequency condition and long low frequency condition. This evidence was that both these conditions have unique practice effects, which disappear when either distance or duration was manipulated to one of the other related conditions (Figures 4, 5 and 6). This suggests that a combination of both duration and distance, possibly defined as velocity, results in the unique effects. The evidence was marginal for two reasons, first that the uniqueness of these two conditions could vanish with additional conditions beyond the scope of the movement constraints imposed here. Second, in the long low frequency condition, ideal duration was achieved, with systematic error, from the first trials and did not deviate and no improvement of the other two variables occurred. Thus the duration perspective was favoured based on task performance across trials for all conditions, with the short high frequency condition results attributed to participants avoiding neurobiomechanical constraints of the ideal duration. Since duration appears to be the controlling variable, inference can be based of this on the neurobiomechanical/experience and feedback continuums. Generally, participants learned the ideal duration faster or easier as driving frequencies decreased. In the short line conditions, there was a distinct change in behaviour from high to medium frequency and in the long the change was from medium to low frequency (Figure 5). In the short conditions, the change was from avoidance of ideal duration to learning to increase movement duration without reaching ideal duration. In the long conditions, the change was from decreasing movement duration to the ideal to having the first movements at ideal duration (Appendix A). These practice effects, particularly the avoidance behaviour, suggest that neurobiomechanical constraints increased as driving frequency increased and experience enabling movements to be easier as driving frequency decreased. When comparing feedback loops against each other for practice effects, over all
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
69
auditory feedback was primary source. In the short line conditions, auditory feedback had greater influence on the lowest frequency movements and haptic feedback had greater influence of the highest frequency movements. With auditory feedback the lower frequency movements had slow increase in the guidance from this form of feedback, where as the high frequency had no guidance (Figure 5). With haptic feedback, distance tentatively appeared to be more stable and less divergent as frequency increased (Figure 4). For the long line conditions, with auditory and haptic feedback improved performance as movement frequency decreased (Figure 4 and 5). The haptic feedback also effected learning ideal distance between the two line lengths. In the lower frequency movements, haptic influenced movement as distance increased by becoming more stable and less divergent from ideal distance (Figure 4). In the high frequency, the reverse was appeared to occur. A visual summery of the practice effects is provided in Appendix A. The practice effects suggest future research could focus on exploring constraints outside of the scope of this study to elucidate the tentative feedback and neurobiomechanical models further. Overall, the systematic deviations and unexpected findings suggest auditory and haptic feedback loops are involved in planning and executing the movement, and these work together on continuums. These continuums are in conjunction with continuums of experience and biological constraints (Appendix A). The planning and execution of movements appears in general to be controlled for ideal duration. Distance appeared to interact with the ability to achieve ideal duration, by enabling people to achieve ideal duration for the longer line and not the shorter. The haptic and auditory feedback was used in all conditions, with auditory being more generally predominant then haptic, indicated by the majority of movement error. The influence of either haptic or auditory feedback depended on main effects of distance and duration, with or without practice effects. Neurobiomechanical constraints increase as driving frequency and line length decreases based on the interaction of duration and distance.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
70
However, based on the effects of practice, the increase in constraints due to decreasing driving frequency appeared easier to improve, indicating the constraints may also increased with driving frequency (Appendix A). The natures of the interactions between the three continuums were complex and difficult to elucidate and simplify based on the current data, thus more research on them would be valuable. Movement Intermittency This section focuses on processing and analysing the results to provide possible explanations and implications for the changes in intermittency due to changes in duration, distance, speed and practice. This process is then followed by an exploratory analysis to determine differences between the two measures of intermittency. Distance and duration. What were the controlling variable(s) that determined the degree of intermittency for the six conditions? There are two perspectives of interpreting the results. The first perspective is represented in Figures 11 and 12, as previously shown by Mathur and Hughes (2009): velocity (or speed) and distance determine the level of intermittency. This implies that the ability to control mean online velocity determines the ability to reduce fluctuations from the ideal speed, which, in turn, is influenced at the same time by the distance travelled. The lack of no main effect of distance is explained by the ideal speed being double in order to maintain the same driving frequency between the two lines, which doubled in length. The two variables covary between the two lines, suggesting that any difference or lack of difference is due to either one or both of these variables. If velocity is the only controlling variable, there should have been a decrease in intermittency in the shorter conditions (Celik, et al., 2009; Doeringer & Hogan, 1998). If the distance was the only controlling variable, intermittency should increase with distance since there would be more opportunity for fluctuations (peaks or amplitude) to occur. Alternatively, distance could have a negative relationship with intermittency, due to shorter movements being more
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
71
difficult than longer movements. However, there was no change between the short and long conditions, indicating that distance countered the effects of velocity and that both variables are key for determining the level of intermittency. The alternative perspective of defining the relationship between the intermittent conditions is that driving frequency or duration was the only controlling variable. This perspective is presented in Figures 14 and 16. For acceleration zero-crossing there was a strong linear model defining the relationship between intermittency and duration as seen in Figure 14. Figure 14 shows that the relationship does not follow the curvilinear relationship observed with driving frequency (see Figure 7). The reason was that the ideal duration for a segment has a linear increase, whereas driving frequency changes duration with respect to one second, in a curvilinear decrease. For normalized jerk-cost (Figure 15), the relationship with duration is monotonically increasing due to the increase in stop periods and stop time (Table 2) as would be expected (Hogan & Sternad, 2009). Exponential function (not currently reported here) and quadratic functions for this relationship had similar r-squared values (Figure 15). The quadratic function was chosen because it fits with the nature of linear quadratic contrast effects. The duration perspective could be chosen because it is more parsimonious (one controlling variable rather than two), and driving frequency is directly guided by auditory feedback from the metronome. This perspective is also congruent with previous theories such as the theory about changes in amplitude (Vallbo & Wessberg, 1993) or the invariant rate scaling model (Terzuolo & Viviani, 1980, with auditory and haptic feedback from the keyboard). However, the unexpected statistically significant shorter duration across the shorter line provides a problem for this perspective. The duration perspective would state that intermittency should reduce in these conditions, which it does not. But, this reduction in duration (and an increase in speed) was proportionally small, and hence may not have been large enough to cause a change in intermittent behaviour. In favour
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
72
160
Mean No. of Acceleration Zero Crossings
140 Y = 12.4669x - 2.099 r2 = 0.955
120
100
80
60
40
20
0 0
2
4
6
8
10
12
Duration (s)
Figure 14. The relationship between the mean number of acceleration zero-crossings and duration for the segments is presented. Each data point represents the segments of the cycles, during a single individual’s trial. In addition, the results of a fitted linear function are displayed.
of this perspective was the finding, previously mentioned, that duration appeared to control the movements in some and between some conditions (Figure 5). However speed could also be a controlling variable in other situations (Figure 6). As Dépeault, et al. (2008) discusses, the problem with choosing either perspective is the covarying nature of duration and speed. Even if the participants were using the additional haptic feedback, Dépeault, et al. found it was due to the space or duration between raised elements that enabled participants to perceive their movement speed. This implies that some measure of time is needed in conjunction with a measure of distance to execute movement at an ideal speed, as might be
Mean Normalized Jerk-Cost
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
3000
73
100
2500
80
2000 60 1500 40
1000
20
500 0
0 0
1
2
3
4
0.0
0.5
1.0
Mean Normalized Jerk-Cost
80000
Plot 1
60000
Y = - 289.36 + 122.54x + 312.17x ^2 r2 = 0.778 40000
20000
0 0
2
4
6
8
10
Duration (s)
Figure 15. The relationship between the mean normalized jerk-cost and duration for the segments is presented. Each data point represents the segments of the cycles, during a single individual’s trial. In addition, the results of a fitted quadratic function are displayed.
expected based on their relationship. This notion is supported by research indicated perception of time has an impact on initiation and execution of movement (Frowein, 1981; Grondin, 2010; Ivry, Keele, & Diener, 1988; Sheriden, Flowers, & Hurrell, 1987; Toplak,
12
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
74
Dockstader, & Tannock, 2006; Wittmann, 2009, 2013). One possible way to distinguish between the two perspectives, duration or velocity and distance, would be to investigate movements when the person has yet to learn the distance or duration of the movement. In these situations the speed could be the controlling variable (Figure 13), but this may not apply to movements after they have learned the duration. Alternatively, they are simply short duration submovement executed to reach the end of the line: combining the ideas from both Terzuolo and Viviani (1980; Vallbo & Wessberg, 1993) and the submovement idea expressed in the introduction (e.g. Henis, 1991). This idea is preliminarily investigated in the next section with practice effects. For the present study duration was chosen as the controlling variable for the duration by distance effects, as it was more parsimonious and was directly guided by the metronome. Practice. What controlling variable(s) decreased the trend in intermittency or was the effects determined by distance alone? To explain this change in intermittency, there are four possible factors influencing the conditions: duration, ideal conditions, and methodology. Duration was a possible factor because it was the controlling variable for the degree of intermittency, therefore could changes in duration across trials also influence the decrease in intermittency? The changes in duration when comparing first to last trial were between 412%. A proportion of the changes in intermittency could be related to changes in duration across the trials. However, the significant proportional change in duration between the long line and short line was between 6-11%, which did not have an effect on degree of intermittency. The difference between these situations was the comparison of between conditions (short to long and suggested to be due to biological/experience constraints) or within condition (first trial, last trial). Was there something unique about the within condition situation? From Figure 13 the first one or two segments of the high frequency conditions were different from the rest of the segments and highly intermittent. This was also
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
75
observed in some of the medium conditions (data can be provided), but could not be detected in the low frequency condition (due to only one cycle being executed in these conditions). The within trial finding for the high frequency conditions was investigated in respect to practice in Figure 16 for the long distance condition. Alicia was chosen for the investigation of practice effects on the first segments, because she had floor effects whereas Simone had not, a finding that supported the earlier assertion that some people could have had floor effects. Making the distinction between someone with floor effects and someone without was important because floor effects across trials might mask or influence any changes within the trials. However this was not the case, since both Alicia’s and Simone’s first segment was still slower then the following segments. These results suggest that during the high frequency conditions people needed to learn the distance or duration, both within trials and across trials. In addition, the results suggest that speed of the movement or short duration submovements could be sources of the intermittency in the first segments. Speed as a factor was supported by the observation (as mentioned in the discussion on task performance) that the high short condition’s uniqueness could be speed, as well as the observation that in the higher frequency conditions participants avoided the ideal duration across trials (Figure 5). However, submovements could have controlled the intermittency in these first segments (see following studies, Terzuolo & Viviani, 1980; Vallbo & Wessberg, 1993). However, if the source was speed this could have come from both distance and duration information and was used at relatively equal levels. At present the results favour that the decrease in intermittency during the high frequency conditions was a by-product of learning the task (as suggested for the medium and low frequency short distance movements in the task performance section) and from a conscious effort to reduce smoothness. The evidence for the conscious effort to reduce smoothness comes from the observation that a reduction in intermittency was the only variable improved with practice in the short high
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
Finger Velocity (cm/s)
Simone
16 cm for 1 Hz
Alicia 80
80
60
60
40
40
20
20
0
0
-20
-20
-40
-40
-60
-60
-80
76
-80 0
2
4
6
8
10
0
2
Time (s)
4
6
8
10
8
10
Time (s)
80
80
60
60
40
40
20
20
0
0
-20
-20
-40
-40
-60
-60
-80
-80 0
2
4
6
Time (s)
8
10
0
2
4
6
Time (s)
Figure 16. Velocity profiles with respect to time, of high frequency (1 Hz) long distance (16 cm) conditions for two participants. The top graphs are the individual’s first trial, and the bottoms ones are their last trial.
frequency condition. Both the duration or speed perspectives would need further research to provide empirical support to determine which was the controlling variable and in which situations. Additional support for the duration perspective comes from the relationship across trials between duration (Figure 5) and intermittency (Figures 8 and 10). This evidence was that the intermittency’s practice effects change with duration across trials in all but some situation. For example, in the short high frequency condition duration decreased and intermittency decreases. The situation where this does not appear to be the case was in the
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
77
long low frequency condition; where intermittency decreases but duration has little change. This raises an issue for the duration perspective, however it was unclear what was changing across trials at the individual level. For example, a majority of participants could have decreased duration and decreased intermittency and a smaller group have an increased duration as they where learning the task. At present, it was unclear what was happening in this condition, as it was also the condition with the sudden decrease in normalized jerk-cost. Thus additional research was suggested to investigate this situation. Ideal conditions provide a simple tentative explanation for the different practice effects between the long line medium (normalized jerk) and low (acceleration zero-crossings) conditions. As mentioned, this suggests a double dissociation. The notion of ‘ideal conditions’ suggests that these conditions were at the ideal duration, distance or speed for practice effects to occur in 18 trials for either measure of intermittency. Speed and or duration could be the controlling variables since they were the difference between the two conditions. This notion was also supported by the task performance analysis, which suggested the practice effects of the general movement in the acceleration zero-crossing condition were speed or duration but for the medium with normalized jerk-cost the controlling variable was duration. Further research on these variables is needed to elucidate possible casual relationships beyond the present explanation. Two methodological variables may explain when the practice effects started to level out and why practice effects levelled out in the medium (for normalized jerk-cost) but not low frequency condition (acceleration zero crossings). These two variables were the duration of trials or number of cycles per trial. In the present study, amount of time was controlled for each trial as was conducted previously by Mathur and Hughes (2009). A low frequency trial took twice as long as both medium and high trials, giving participants different time periods practise the different driving frequencies. This method appeared to be conducted to enable a
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
78
complete cycle of the lower frequency movement. However, in terms of number of cycles or opportunities to practise a cycle, the low condition had 18 cycles for all the trials, the medium condition, this had 18 cycles in nine trials, and high frequency had 18 in just over one trial. The majority of previous research used the exact number of cycles per trial during learning tasks or measuring behaviours (Cooper, et al., 2007; Doeringer & Hogan, 1998; Fradet, et al., 2008a; Schneider & Zernicke, 1989), which makes it difficult to compare directly trial to trial with other studies. However, the trends in data from this study can be compared between the conditions and can be compared, these studies. In addition, this difference may suggest a possible avenue for future research to determine which variable, duration or count, was controlling for the effects of practice. Summarizing the movement intermittency and task performance analyses, the relationship between duration (provided by auditory feedback), distance (provided by haptic feedback) and velocity on changes in control represented by intermittent was dependent on the situation. From the intermittency data, it appears that duration was the controlling variable of intermittency. This evidence comes from both the effects of duration and distance with and without practice effects. Similar to the task performance analysis, there was still the possibility that speed was involved to a lesser degree. Comparing with the task performance analysis, one difference was; when taking the perspective that the ability to reduce intermittency was an indicator of lower neurobiomechanical constraints compared to conditions that do not have a decreased. Then this perspective would indicate that the short high frequency movements would appear to have lower constraints then short lower frequency conditions, opposite to what was observed in the task performance data. Combining with the task performance data, the effects of auditory and haptic feedback appear to be on continuums, which interact with experience, neurobiomechanical constraints or both (Appendix A). Alternatively, there could be a threshold between some of the conditions
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
79
where the nature of control changes. A continuum perspective was suggested to be the case for this data, based on the idea that the movement requirements were difficult and the driving frequency across the distances provided novel sets of forces (Shadmehr & Mussa-Ivaldi, 1994; von Hofsten & Rönnqvist, 1993) from an internal locus of control. Acceleration Zero-Crossings or Normalized Jerk-Cost An exploratory analysis of differences between the two measures of intermittency, acceleration zero crossing and normalized jerk-cost, was performed. According to visual analysis (Table 3 and 4) an interesting finding emerges for the long medium and low conditions; there appears to be a double dissociation between the measures of intermittency and practice effects. This double dissociation was supported by differences in the statistical tests; significant interaction between distance and duration; different contrast effects; and the significant complete interaction for normalized jerk-cost compared to the acceleration zerocrossings. Combined the different results either measure support previous research that there are differences between jerk and peak based measures of intermittency (Hogan & Sternad, 2009). Thus, the two measures might measure different components of intermittency. The differences have theoretical implications and are important to investigate further. Two key findings that supported differences in the two measures, based on the measures’ representations of velocity profiles, for example profiles see Figure 17: two segments with the same number of acceleration zero-crossings can have marked different values of normalized jerk (Figure 17A); an increase in one measure does not mean an increase in the other (Figure 17B-D), even though they both are aimed at measuring the peaks in velocity (Rohrer, et al., 2002). These two findings were found in two situations: the difference between the peaks and valleys (amplitude) increased (Figures 17A and 17D) and one peak was larger than the rest (Figures 17B and 17C). These results imply that measures are different underlying factors. It is important to note that these findings were from a brief
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS. A
30
Finger Velovity (cm/s)
B
15
AZC: 28
20
NJC: 1095.17
10
10
5
0
0
-10
-5
-20
80
AZC: 109
NJC: 9314.32
AZC:94
NJC: 11863.77
-10 AZC:28
-30 4
6
NJC: 2167.63 8
10
12
14
16
-15 2
18
4
6
Finger Position (cm)
Finger Position (cm)
C
15
D
15
AZC: 75
8 10 12 14 16 18 20 22 24 26 28
NJC: 6020.32
10
10
5
5
0
0
-5
-5
AZC: 97
NJC: 17155.16
AZC:95
NJC: 33772.89
-10
-10
NJC: 12927.02
AZC:71 -15
-15 4
6
8
10
12
14
Finger Position (cm)
16
18
2
4
6
8 10 12 14 16 18 20 22 24 26 28 Finger Position (cm)
Figure 17. Examples of velocity profiles for comparing acceleration zero-crossing estimates (AZC) to normalized jerk-cost (NJC) means. Profile A was under the medium diving frequency (0.25 Hz) and the others were under the slow driving frequency (0.0625 Hz). The values above the dashed line are for the associated positive velocity segment above the line and the same applied below the line. The grey square represents the location of the horizontal raised line with vertical lines at 1cm intervals.
investigation, and thus other differences in the measures could exist such as stop periods and stop time. Future research would be needed to determine any additional practical differences. Overall, even though normalized jerk measure was designed to measure the peaks and duration between them (stops and valleys), a primary contributing factor to the normalized jerk value appears to be amplitude (a measure of duration), which can easily override the number of peaks. This observation indicates that acceleration zero-crossings tend to be more
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
81
sensitive for changes in peak count than normalized jerk, whereas normalized jerk enables the measure of both amplitude and number of peaks. This analysis tentatively suggests that the difference in movement output between the long medium and low frequency conditions was improvements in peaks for low and improvements in amplitude for medium. Thus, the data presented in this section and the results section indicate that both measures should be used, together or independently, and understanding the differences between the measures is important. Effects of Speed on Intermittency This section focuses on incorporating the results of the present research with the literature on changes in speed, covering studies with similar speeds of movement and theories of control. The present results support and add to both the studies suggesting that slower movements become smoother with practice (Darling, et al., 1988; Schneider & Zernicke, 1989). Although difficult to assess, based on the other studies’ descriptions, the slowest movements in both studies appear to be slightly faster than those for the medium frequency, and slower than those for the high frequency conditions which were used here. Although the present study showed improvements in the high frequency conditions with fast movements, as mentioned, this could be due to learning the task while blindfolded and using auditory feedback. This study adds to the speed and intermittency research with the effects of duration, distance, haptic feedback, auditory feedback, movement type and slower movement speeds. The results, however, appear to conflict with those results from the study by Doeringer and Hogan (1998) who found no practice effects with their constant velocity task. The possible conflict with Doeringer and Hogan’s study may not exist since there are number of distinct differences. The slow condition in the constant velocity task in their study was 10°s for 10 seconds. This was similar in duration and profile of velocity to the medium
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
82
frequency conditions here, expected to be eight seconds a segment. The differences between the conditions were in distance, method and measurement. Based on the results of this study, distance plays a role in the ability to improve intermittency, which could explain the difference between the studies. However their measurement was relative and thus difficult to compare distance since this study involved an absolute distance. Their method also involved grouping trials by speed, having the fast trials first in order to have practice effect happen as fast as possible, and some trials had no vision feedback while others had. In addition, they used a different measurement of smoothness, coefficient of variance, and different measures of intermittency, all leading to obtaining different results (Hogan & Sternad, 2009; Rohrer, et al., 2002). At this stage the most likely reasons seems to be the difference in distance travelled or measure of intermittency. Overall, the results of this study are congruent with previous studies when differences in distance, speed and intermittency measures are considered. The results of this study can be further incorporated into the theories of control of slow movements. As mentioned there are two general positions: control changes as speed decreases (Fradet, et al., 2008a; Wisleder & Dounskaia, 2007) or control remains unchanged (Flash & Hogan, 1985; Milner, 1992). The two factors that could affect control in the present study were distance and duration. The velocity profiles indicate that the movement profile were similar across distances. There was no systematic difference between distances across the driving frequencies on acceleration-zero crossings or normalized jerk-cost. This suggests that the method of control did not change between the two distances. Yet, there was a distinct difference between the high frequency movements and the lower frequency movements. The high movements appeared to be more ballistic, but the lower movements were not. However, if submovement theory is considered in explaining the slow movements, then each ‘submovement’ was ballistic and fast movements are at least one submovement
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
83
(Flash & Hogan, 1985; Terzuolo & Viviani, 1980; Vallbo & Wessberg, 1993). The finding suggests that submovement control, if used, could be the same between the high and low frequency conditions. In combination, control did not change with distance but it could have changed with driving frequency. Thus the profiles could support either position. There has already been reasonable evidence that components of movements do not change proportionally to total duration (Gentner, 1987). However, the components at the global level of movement that are reviewed by Genter are discretely or functionally different; for example the support phase and swing of walking. Thus this research may not be directly applicable to the research on fluctuations in velocity with a single functional movement. The present study’s results from the task performance and the practice effects of intermittency can be combined with other research to support a single theory that combines both perspectives, similar to the idea proposed by Gentner (1987). This single theory follows the idea that there is one method of control (Flash & Hogan, 1985; Milner, 1992; Thoroughman & Shadmehr, 2000) which becomes less efficient as speed decreases due to a combination of biological constraints (Fradet, et al., 2008a; Wisleder & Dounskaia, 2007); noise (Celik, et al., 2009; Nagasaki, 1991); and intermittent signal (DeLong, 1972, 1990; Pearson, 1972). The tendency for people to avoid slow long duration movements (van der Wel, et al., 2010) suggests one control mechanism optimized for fast movements becoming less effective as requirements decrease speed. Accordingly, neither minimum-jerk or minimum-sap models are effective for predicting slower movements (Wiegner & Wierzbicka, 1992). The results of this study found participants could move slowly for a long duration if asked to, but the movements became more intermittent. In addition, participants had difficulty with the shorter distance movements regardless of duration, and distance had an effect on practice. The difficulty with short movements increased as driving frequency increased. This study distance as an important variable to the one control method described.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
84
Therefore, a single control mechanism for arm movement is suggested. This mechanism becomes less efficient as movements are executed for longer durations, and shorter distances or shorter durations and shorter distances (Appendix A), with shorter movements being harder to control than longer movements, and easer to control for longer durations. Hence, using speed to define a movement may not be a useful scientific description for studies with a similar design; an alternative term could be movement frequency, timing or duration and specify the distance to be travelled. Movement frequency (or the frequency of movement) is suggested as it defines the duration between two locations, and the description is more intuitive and simple to use, compared to the movement’s duration. Thus research should focus on the effects of distance and duration before approaching the effects of speed. This perspective does not remove the usefulness of speed. Defining the movement frequency and using it as the primary defining variable with distance in movement research, has advantages. A primary advantage was that it provides scope for the movements. For example, if one study uses the duration of one cycle of movement as the unit, then another study measures segments of the cycle or submovements as the unit. Then the results of both studies will be easy to compare, assuming the studies’ designs are similar enough. If the comparison was between one study with speed and the other with movement frequency and distance or speed then it becomes more difficult to compare. On a more practical perspective a reader can obtain the speed, or a close approximation, from the study with movement frequency and distance, whereas the reverse cannot be done with the study using speed as the defining variable. In a general sense both studies are measuring the same thing in slightly different manner. There is also a general advantage as it has been commonplace within studies, particularly studies that focused on constant velocity, to define the duration and distance (Milner, 1992; Morgan, et al., 1994; van der Wel, et al., 2010; Wisleder & Dounskaia, 2007). However, it appears less common to investigate the effects of
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
85
either variable, only one study (Abend, Bizzi, & Morasso, 1982) was known to the author’s knowledge. Thus it is suggested that these two variables could incorporate into future research on, or in addition to, the speed literature on movement. Practice Effects and Haptic Feedback This section focuses on integrating the results with the literature on practice and feedback. The present study was unique in its use of haptic feedback. Based on the previous interpretations of the data, the participants use haptic feedback and it restricted their movement in the short line condition resulting in the timing issue. This raises the question of what role did the additional haptic feedback of the vertical bars play above what was provided by the horizontal line? This additional haptic feedback could be a source of intermittency, a factor in decreasing intermittency that occurred with practice or a combination of both. The vertical bars can be a source of intermittency due to the strategy used for the execution of the movement or friction: participants could be, in some situations, executing submovements from bar to bar in time with the beat. This strategy might be observed in the movement right to left (negative velocity) of Figure 17A and 17D. The seven peaks in Figure 17A represent a movement per beat per bar or per gap (space between bars), suggesting each peak were deliberate timing related submovements, rather than intermittency resulting from the method of execution when aiming for a constant velocity. Although this strategy could be used, there was no condition where practice increased normalized jerk-cost and decreased acceleration zero-crossings. This observation indicates that if the strategy was used, then its use was reduced with practice. The rationale behind the friction or structure of the surface was that the raised elements of the bars increased friction and interfere with the movement of the finger (Hughes, et al., 2011 with braille dots), assuming that making contact with the bars interferes with movement across the surface and increases intermittency. Two studies have
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
86
investigated a relationship between surface structure and tangential fluctuations (Hughes, et al., 2011; Smith, et al., 2002). Smith, et al. (2002) found that amplitude of the fluctuations in force was related to longitudinal spatial period, and that frequency was related to velocity and spatial period. In addition they found support for surface texture or type of surface impacting the level of tangential fluctuations. In contrast, Hughes, et al. (2011) found that increasing the number of dots in a braille cell had little change in fluctuations. The surface in this study was similar to that used by Smith, et al. (2002), since both had a plastic surface, larger spatial period and bars, although their attributes were still different from this study. Their bars were made of dots and their surface was hard plastic, where as here the bars were straight and made of softer plastic. Overall, the present research indicates that the additional raised bars could have increased the intermittency in this study, but further research is needed to elucidate this possible theory. When comparing this study to previous studies, the effects of the vertical bars on the ability to increase the smoothness in the condition could be a factor for all, some or none of the improvements found. The previous studies investigated the practice effects of visually guided fast movements to a target (e.g. Elliott, et al., 1999; Pratt & Abrams, 1996). These studies indicated that without visual feedback participants’ movements became smoother with practice (e.g. Khan & Franks, 2000). During these conditions the improvements could be due to proprioceptive feedback, procedural memory, mental representation, limited amounts of haptic feedback or some combination of these. The main differences between this study and the nonvisual conditions in the previous studies were speeds involved, sight of the start and target of the movements before moving and accuracy/task requirements (e.g. Khan & Franks, 2003; Pratt & Abrams, 1996). Two perspectives are explored when comparing this study to these previous on-visual conditions. One perspective could be that these differences are minimal. The slower speeds in
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
87
this study made it easier to improve smoothness of movements, due to more opportunity during slower movements to alter trajectory and speed across the line. The ability to see the start and end location was important for visually guided movements (e.g. Proteau & Marteniuk, 1993), and thus enable the possibility of a mental representation of where the hand was during the movement. The haptic feedback of the horizontal line in this study could guide the movements to end of the line. In addition, as mentioned, the first segment(s) were slower conditions, which could enable participants to create a mental or procedural representation of the line to guide the later cycles (Figure 13). This principle could also be the case across trials (Figure 16). The representation of the line created by the participants could be built with a combination of duration, distance and relative proprioceptive feedback (e.g. movements of the wrist relative to the elbow). The type of accuracy required was fast as possible to a target in the previous studies where as in this study it was smooth constant velocity and end of the line as the target. These differences indicates a problem with this perspective since ‘fast as possible’ movements might be controlled differently than movements that aim for a specified constant velocity, other than just the changes in speed. The closest study so far involved a timing constrained task, which found practice had more influence on the accuracy of non-visually guided movements from start location to target than the visually guided movements (Proteau, et al., 1987). Overall, this perspective suggests that haptic feedback from the horizontal line could be enough to enable practice effects without the bars. The baseline condition, without the vertical bars, could help researchers to determine the degree of contribution of the additional haptic feedback from the bars. An alternative perspective could suggest that differences between previous studies’ no-vision conditions and this study’s conditions are distinctly different. However, the findings can still be compared in terms of the availably of feedback. Visual feedback does not appear to have an impact with moderate or small amounts of practice (Pratt & Abrams,
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
88
1996), but will interact with practice after extensive practice (Khan & Franks, 2000). If we assume that participants in their studies and this study have more experience with visually guided movements to a target than tactile guided constant velocity tasks, then it would likely require more practice than the 18 to 144 cycles in the current task for people to learn to use the feedback from the bars above that provided by the straight line. This is likely, because Khan and Franks (2000) found consistent trend in data between 750 and 1,250 movements (depending on measurement variable) where visual feedback was having different practice effects on intermittency then the non-visually guided movements. In addition, the horizontal line could still provide other forms of information about the distance of the line including proprioceptive feedback (receptors from the shoulder to the tip of the finder), procedural memory and a mental representation of the line that could enable practice effects to occur. Overall, this perspective suggests that maybe these bars will be useful after extensive practice. Future research should extend this paradigm with a baseline condition and additional cycles or trials to determine the effects of haptic feedback from the bars. In summary, at present there appears to be too many possible variables influencing the control of the movement and these variables need to be controlled to detect effects from the bars. Cognitive Sources of Intermittency This section focuses on integrating the results with the literature on the cognitive causes of intermittency. The present study supports the application of some form of cognitive theory explaining a proportion of the intermittency found during movement of the arm. The practice effects indicate that people can consciously improve intermittency of arm movements guided by the finger. Suggesting that participants were able to improve their movement, using feedback, planning of the movements (feedforward model), execution of the movements or a combination. First, it is worth acknowledging that the improvement participants made in
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
89
performance may not of come from conscious control over their movements. It was distinctly possible that any improvement could of occurred automatically. The current study paradigm would possibly need to be changed considerably for a control condition where conscious control of movements or reduction in intermittency cannot occur. A simple example would be repeating the current design but without the request for smooth movement. However, the request for continuous movement could confound this, as moving continuously would reduce the intermittent stops. Alternatively, a design could be based on van der Wel, et al. (2010) study with the addition of reinforcement for increases in smooth movement. This previous idea would have it’s own complications, such as people becoming aware of what they need to do for reinforcement. An alternative argument from a practical position; is that the majority of movements with a low driving frequency (painting, surgery) that require learning smooth execution are learned with the conscious awareness of the need for smoothness. Braille could be an exception to this, as it has yet to be determined if conscious improvement of intermittency is functionally useful to become an efficient reader. Future research could explore study designs to find an effective control condition. It was suggested that the participants were using some conscious control in this study to reduce intermittency. This was based on idea that if improvements in intermittency were purely automatic, external feedback driven, then there would not of been learning in the short high frequency or the long low frequency conditions. In the short high frequency was the avoidance behaviour, indicating the focus was learning to smooth movements and not to match the duration set by auditory feedback. In the long low frequency there was little or any change in duration as it was close to ideal while learning to smooth movements still occurred. In the other conditions there was no direct indication that conscious effort was occurring. However, indirect support comes from acknowledging that trials were randomised and if conscious effort were occurring in a third of trials and not in others then participants would
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
90
have to be switching between conscious and unconscious effort. This switching was certainly possible but currently appears unlikely, which tentatively suggests conscious effort was occurring to reduce intermittency. If there were no conscious effort in the other conditions, then it would indicate that amount of conscious control was less overall. Alternatively, the conscious effort was dedicated to learning the task and not reducing intermittency, as previously suggested in the short lower driving frequency movements. This study adds to the research that visual feedback is not the only cause of intermittency (Doeringer & Hogan, 1998; Hughes, et al., 2011). However, the fluctuations could be explained as corrective in nature, in this case speed correction rather than end point error correction since the primary target was smooth speed. Combined with the negative curvilinear relationship found between velocity and intermittency, these results support Dounskaia and colleges’ claim that decreases in speed are a source of intermittency (Fradet, et al., 2008a; Wisleder & Dounskaia, 2007). The difference in the aim of the task may be key in differentiating between endpoint error correction and speed error correction. If the only aim was to arrive at the end of the line in time with the beat, then movements would have likely changed (van der Wel, et al., 2010) compared to other tasks that required constant velocity (Doeringer & Hogan, 1998; Morgan, et al., 1994). However, the results of this study still exclude visual feedback as the primary source of intermittency suggested previously (e.g. Elliott, et al., 1999; Meyer, et al., 1988). These results thus do not support the closed-loop visual feedback theory. Instead there appears to be in this study a combination of a closedloop haptic feedback loop and an auditory feedback loop (Appendix A). In this situation, the haptic loop informed the participant of distance information and the auditory loop informed the participant of time information as with other studies that use an internal beat (move as fast as possible) or external beat (a metronome). Based on the analysis of the intermittency and performance data, the perception of time was the dominant controlling (Figure 5, 14, 15)
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
91
variable (the imitation and execution of behaviour), and this interacted with the distance information as it became easier to perceive in the longer duration movements or became constrained by the shorter distance. For example, Dépeault, et al. (2008) found temporal and special cues processed from haptic feedback. The feedforward model summarized by Thoroughman and Shadmehr (2000) was based on trajectory planning following the minimum-jerk model (Hogan, 1984). The normalized jerk-cost failed to distinguish the decrease in velocity peaks detected by the acceleration-zero crossings. This in turn suggests the model by Thoroughman and Shadmehr (2000) needs a new model for planning trajectory if it is to account for slower movements such as snap (Wiegner & Wierzbicka, 1992). Alternatively, another different internal model could be used (Kawato, 1999; Mussa-Ivaldi, 1999). If the theory that the cognitive strategy that controls movement becomes less effective due to neurobiological sources of intermittency, then this model may still be applicable with feedback loops that are used to reduce the intermittency. The submovement theory (Henis, 1991; Milner, 1992; Milner & Ijaz, 1990) as a method of execution provides a possible explanation for the results in this study. Due to the simplicity of the submovement theory, it was tempting to try and fit the results to the theory, especially in view of the different results of jerk and acceleration zero-crossings. The different results suggesting a differentiation of a blending control mechanism (initiation of submovement) and submovement control (execution of submovement), similar to the motor issues resulting from damages to different parts of the cerebellum (Ivry, et al., 1988; Wittmann, 2009). However, the results do not completely fit, nor do the results disprove the theory either. In Figure 8A, the smoother left to right movements with lower jerk could be explained by this theory as being a blended version of movements going right (Henis, 1991). Alternatively, as mentioned, another cognitive strategy can explain these peaks as deliberate
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
92
timing related submovements. In addition, consider the velocity profiles in Figure 3, particularly the lower driving frequencies. Visually some of the peaks look like distinctive submovements during the lower movements, but some of the smaller or flat ones do not. These flat peaks have been described with computational models as submovements (e.g. Rohrer, et al., 2003). Computational models are useful because they indicate a way the brain could perform the task, but the models do not infer that the model is how the brain does execute the movements. With extensive testing, however, some confidence maybe obtained. Further support for the submovement theory (Milner & Ijaz, 1990) comes from the frequency of stop and dwell periods which indicates that submovements occurred more often as driving frequency decreased. However, this does not mean all of the fluctuations were submovements. If the work by Terzuolo and Viviani (1980; Viviani & Terzuolo, 1982) is included then the possibility arises for future research to investigate level of skill: after extensive practice with this task beyond the quantities of practice by Khan and Franks (2000; Shadmehr & Mussa-Ivaldi, 1994), the person will become highly skilled, and a minimum number of submovements is executed by the person based on the ratio for the duration of the movement (Terzuolo & Viviani, 1980; Viviani & Terzuolo, 1982). Thus, submovement theory can explain the intermittency that was cognitively penetrable, and it could also explain some of the fluctuations that are not cognitively penetrable because once the ratio is achieved the submovements cannot be consciously decreased. Additional research is needed to determine and explore how the evidence for biomechanical source(s) merges with the submovement theory. In summary, the results support the use of a cognitively penetrable source for intermittency. Either an internal model or an executive model based around submovements could apply, and as mentioned these models are not exclusive and this model then interacts with different feedback, visual, haptic, auditory and likely proprioceptive. Based on the
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
93
previous research the submovement theory is a strong possibility. In addition, this theory works well with the single control mechanism described for movements that become less efficient due to biological constraints, system noise and intermittent signals. Biomechanical Sources of Intermittency The data support the idea that a proportion of intermittency during the movement executed by the participants is from biomechanical sources. However, they do not at present support either an intermittent signal or noise source. To provide some indication for either source, a preliminary single subject design study with Fast Fourier Transform (discrete version) was conducted. From the main group, three subjects were chosen, on the basis of the visual analysis of their power spectrum data (from MovAlyzeR of the whole trial) indicating these participants were an exemplar of commonly occurring bandwidths. Two of the three subject’s data have been previously displayed. One was female who was referred to as Simone (19, right index and ambidextrous) and two males Kryten (24, right index and right handed) and Niko (18, right index and right handed). Power spectrum analysis (a type of Fast Fourier Transform) was conducted (MATLAB R2009a) on acceleration (second motion derivative) for the segments of the first and last trials for each person. Power spectrum analysis detects the occurrence of cycles in the acceleration profile at different frequencies (Vallbo & Wessberg, 1993). The quantity or amount of the cycle at a specific frequency is represented by power (Ingle & Proakis, 2012). This analysis was intended to show how often different frequencies of cycles occur in acceleration. To reduce noise and increase the number of segments (increasing power and the ability to detect meaningful frequencies), the comparison for medium driving frequency was conducted with three trials and the low frequency was conducted with four trials. For visual analysis, a peak in frequencies was the point between an increase in the power of the frequencies and a decrease that is relatively
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
94
higher power than the frequencies around it. The peaks can be in a bandwidth or range of frequencies. For a single subject a plateau of peaks or a wide distribution of similar power near the top of a peak represents a bandwidth. Across participants this bandwidth can be a combination of peaks or bandwidths from each person. The results for the fast frequency and medium conditions are presented in Figures 18 and 19, and for the low frequency in Figures 20 and 21. The results provide extensive information and could be analysed for many factors, such as investigating the amplitude (Vallbo & Wessberg, 1993). For simplification the following analysis focuses on two main topics for these individuals: frequency bandwidths and practice effects. 8 cm: Last Trials
8 cm: First Trials 5e+7
5e+7
Driving Frequency: 1.00
Driving Frequency: 1.00
1.2e+7
4e+7
1.2e+7
4e+7 8.0e+6
3e+7
8.0e+6
3e+7
4.0e+6
2e+7
4.0e+6
2e+7
0.0 0
2
4
6
0
1e+7
Power Spectrum
0.0
8 10 12 14 16 18 20
2
4
6
8 10 12 14 16 18 20
1e+7
0
0 0
2
4
6
8
10
12
14
16
18
20
0
2
4
6
Freqeuncy (Hz) 1.2e+7
Driving Frequency: 0.250
1.2e+7
8.0e+6
8.0e+6
4.0e+6
4.0e+6
0.0
0.0 0
2
4
6
8
10
12
Frequency (Hz)
8
10
12
14
16
18
20
Frequency (Hz)
14
16
18
20
Driving Freqency: 0.250
Simone Kryten Niko
0
2
4
6
8
10
12
14
16
18
20
Frequency (Hz)
Figure 18. Power Spectrum analysis of the 8 cm high (1 Hz) and medium (0.25 Hz) frequency conditions for three participants. The small graphs in the high conditions are a scaled version of the associated condition so that the ordinal scales match the medium conditions. Note: the high conditions have two trials and medium have three.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS. 16 cm: First Trials
16 cm: Last Trials
Driving Frequency: 1.00 1.2e+8
Drving Freqeuncy: 1.00 1.2e+8
1.6e+7
1.6e+7
1.2e+7
8.0e+7
1.2e+7
8.0e+7
8.0e+6
8.0e+6
4.0e+6
Power Spectrum
4.0e+6
0.0
4.0e+7
95
0
2
4
6
0.0
4.0e+7
8 10 12 14 16 18 20
0.0
0
2
4
6
8 10 12 14 16 18 20
0.0 0
2
4
6
8
10
12
14
16
18
20
0
2
4
6
Freuquency (Hz) Driving Frequency: 0.250
1.6e+7
8
10 12 14 16 18 20
Frequency (Hz)
1.2e+7
1.2e+7
8.0e+6
8.0e+6
4.0e+6
4.0e+6
0.0
Drving Freqeuncy: 0.250
1.6e+7
Simone Kryten Niko
0.0 0
2
4
6
8
10
12
Frequency (Hz)
14
16
18
20
0
2
4
6
8
10
12
14
16
18
20
Frequency (Hz)
Figure 19. Power Spectrum analysis of the 16 cm high (1 Hz) and medium (0.25 Hz) frequency conditions for three participants. The small graphs in the high conditions are a scaled version of the associated condition so that the ordinal scales match the medium conditions. Note: the high conditions have two trials and the medium ones have three.
Four observations are mentioned in regard to the frequency distributions. The previous studies found a band of frequencies between 8-10 Hz (Vallbo & Wessberg, 1993) 69 Hz (Groß, et al., 2002) at the fingertip. There appears to be bandwidths between 7-10 Hz and 3-5Hz that depends on participant and condition. Both bandwidths across participants appear to occur together and independently, suggesting two different processors. The second low frequency bandwidth has a similar range to slow resting arm tremor of 4-5 Hz from cerebellar damage (Findley, 1988 as cited by Doeringer & Hogan, 1998), and both low bandwidths are connected with stationary and voluntary movement for Parkinson’s disease
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS. 8 cm: First Trials
8 cm: Last Trials
Driving Frequency: 0.0625
3e+6
3e+6
2e+6
2e+6
1e+6
1e+6
Simone
0
0 0
2
4
6
8
10
12
14
16
18
0
20
2
4
6
3e+6
3e+6
2e+6
2e+6
1e+6
1e+6
0
0 0
2
4
6
8
10
12
8
10 12 14 16 18 20
Frequency (Hz)
Frequency (Hz)
Power Spectrum
96
14
16
18
Kryten
0
20
2
4
6
8
10 12 14 16 18 20
Frequency (Hz)
Frequency (Hz) 3e+6
3e+6
2e+6
2e+6
1e+6
1e+6
Niko
0
0 0
2
4
6
8
10
12
14
16
18
20
0
2
4
Frequency (Hz)
6
8
10
12
14
16
18
20
Frequency (Hz)
Figure 20. Power Spectrum analysis of the 8 cm low frequency (0.0625 Hz) conditions for three participants. Note: Each condition had four trials.
(Helmich, Hallett, Deuschl, Toni, & Bastiaan, 2012). In addition to both these two bandwidths, the high conditions have an even lower bandwidth around 2 Hz (Figure 18 and 19) as would be expected from the acceleration profile for these conditions (Figure 16). The two higher bandwidths in the high condition are probably from the first segments and
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS. Driving Frequency: 0.0625 8e+6
16 cm: First Trials
8e+6 6e+6
6e+6
4e+6
4e+6
2e+6
2e+6
0
97
16 cm: Last Trials Simone
0 0
2
4
6
8
10
12
14
16
18
20
0
2
4
6
Frequency (Hz)
8
10
12
14
16
18
20
Frequency (Hz) 8e+6
8e+6
Power Spectrum
Kryten 6e+6
6e+6
4e+6
4e+6
2e+6
2e+6 0
0 0
2
4
6
8
10
12
14
16
18
20
0
2
4
6
Frequency (Hz)
8
10
12
14
16
18
20
Frequency (Hz)
8e+6
8e+6
6e+6
6e+6
4e+6
4e+6
2e+6
2e+6
Niko
0
0 0
2
4
6
8
10
12
14
16
18
20
0
2
4
Frequency (Hz)
6
8
10
12
14
16
18
20
Frequency (Hz)
Figure 21. Power Spectrum analysis of the 16 cm low frequency (0.0625 Hz) conditions for three participants. Note: Each condition had four trials.
fluctuations in the following segments. There also appears to be a bandwidth around 2 Hz in some of the low conditions, but the power was generally lower than the other two bandwidths (Figure 20 and 21). This bandwidth was similar to the 1~2 Hz that was found by Doeringer and Hogan (1998). Together, the three bandwidths appear to be associated with different functions the signal for initiation muscles for the movement/submovements (2 Hz see the
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
98
high frequency conditions), isometric intermittency or tremor (3-5 Hz) and execution (7-10 Hz) of submovements. It could be distantly possible that the 7-10 Hz controls the tremor as it might expect this to be the highest frequency. This idea has an issue with the medium condition, as the 2 Hz bandwidths are very noisy. There were other bandwidths and peaks but these were not common and appear to depend on the individual such as Simon’s movements in the last trials of the 16 cm low frequency condition. When the mediums to the low conditions are compared, the power decreases and number of peaks increases considerably. Lower power can occur from less segments, shorter duration of the acceleration cycles, and a decrease in amplitude. The increase in number of peaks has a number of possible causes that include: the noise from lower number of segments, biomechanical system noise or an increase in number of bandwidths. The first cause was unlikely because there are only four additional segments in the medium condition. In addition although not reported here, if the extra segments in the medium condition were removed the distributions of peaks maintained their smoothness. Thus a combination of the other two causes is suggested. Overall, there appears to be large differences between subjects and some conditions have no discernable peaks, which could be attributed to general biomechanical system noise or an increase in number of bandwidths. The practice effects are specifically noted in the high frequency conditions with an increase in power for the peak at 2 Hz for Simon and Niko. Since the duration of the signal reasonably fixed (Figure 16) the increase in power is suggested to be from amplitude. The Figures 18-21 also depict that there were changes in the peaks from the first trials to the last for the medium and low conditions. These changes, particularly for the medium and low condition, should be given less weight, because the changes (or a proportion of them) could be due to within trial variability for participants. To investigate this further, additional trials or trend analysis across the trials are suggested.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
99
In summary, results can be combined within the theory of a single control mechanism; the control mechanism becomes less efficient due to system noise and at least two intermittent signals, represented by the two higher bandwidths. The control of the movement with the two signals may be cognitively penetrable based on the within subject and within trial differences. The three signals could be related to different aspects of the submovement execution process. As mentioned this single subject study was a preliminary investigation and thus would require future research to generalize the current results. Applications and Future Research There are many applications of this research, the following section focuses presenting ideas that may or may not be worth perusing. The majority of these ideas would require additional studies. These additional studies are needed to adapt the results of this study to the environment. One example could be finding out if the degree of practice effects with haptic feedback has practical utility. The degree of decrease in intermittency due to practice in this study was around 10-30% depending on condition. Whether or not this reduction has practical utility is dependent on the area of application and number of trials. For example with surgery, depending on how long the practice effects last for, doing a task that reduces the smoothness of a surgeon’s movements by 10-30% before an operation might be useful, whereas for braille readers a reduction of 10-30% for a short period may not be considered of much practical use. However, in both situations amount of practice could be important for the movements, especially since the low frequency movements have yet to start levelling out. These movements (as mentioned) with the current task, could be considered reasonably hard and would need extensive practice to adapt to the constraints set by the CNS (an internal locus of control), to see the full practical improvements achievable for braille readers or surgeons. One way to test this would be to compare efficient braille readers or surgeons to beginners on the intermittency of their movements during their specific task. If future
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
100
research determines that the ability to improve the movements by 10-30% is not practically useful (and the degree of intermittency does not reduce any further), the reduction would still be relevant for understanding the control of movements. What about a movement involving the whole body, for example the job of dolly grips? The combined research indicates shorter frequency (or timing) movements can become smooth but the speed is less constant which is affected by different sources of feedback, however longer frequency movements are executed around the constant speed but are not smooth. Practice effects indicates that a proportion of the fluctuations can be reduced due to cognitively penetrable sources of intermittency, neurobiomechanical constraints however, appear to increase as movement are executed over short distances and short frequencies or long distance and long frequencies. To take these findings and apply them to work of a dolly grip could be useful for training purposes and could make current dolly grips more effective. More research would be needed specifically in their environment, because the whole body is involved not just the arms, as well as other factors such as the weight of the dolly, the surface the dolly is traveling on and the type of strength of the individual. Determining the effects of the haptic feedback from the raised bars in this study would open up applications for populations such as braille readers, surgeons, seniors, and stroke recovery patients as mentioned in the introduction. This type of haptic feedback could play an important role, for example with seniors, creating a procedural technique that they can be used to improve or maintain their control of movements. With braille readers, determining the effects of haptic feedback could be used to help people learn braille. However, the fluctuations may enhance the ability to read braille. This idea is based on two concepts: tangential forces are related to and maybe a determinate of the perception of roughness (Smith, et al., 2002) and an increased in perfection of the roughness of raised braille dots may improve the ability to read braille. These are just a few of the possible
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
101
applications that the ability to improve the smoothness of movements with feedback can be applied to and future research can further elucidate. There are two main areas with which future experiments can expand on the paradigm used in this study: One possible area is pre-experiment manipulations, such as choice of participants and type of instructions given, for example, comparing blind people to sighted people to determine if blind people have an advantage. The other area involves the conditions during the experiment, such as number of trials, frequency of vertical bars, surface texture, length of the lines, order of trials, number of participants, reinforcement and intertrial knowledge of results. And various analytical techniques can be used, such as computational modelling, cluster analysis and interobserver agreement. Many of these manipulations have being used in the research so far and combined with these other paradigms can be used to create the optimal situation for improving the ability to execute movements as smooth as possible. Alternatively, these manipulations could be aimed at finding out under what conditions movements cannot be or are harder to be smoothened. From both aims, useful information can be obtained about the control of movements and aspects of touch perception or haptic feedback. Other avenues for future research exist; for example exploring the difficulty with short movements. A few participants mentioned that the short low frequency condition was the hardest and other participants mentioned other conditions. Observed improvements in difficult tasks, such as threading a needle, would suggest that we could improve our ability in skills that require moving for short periods with precision over short distances. Are the challenges that people face when being required to make these short movements related to control issues, biomechanical issues or both? If people can improve these movements with enough practice it might imply both as discussed above. The present study can be combined with avenues of research on movement control and intermittency and the potential is open
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
102
wide for finding practical utility. Perception of Time and Parkinson’s Disease This section expresses my current opinion on the research to date and includes all my references and likely sources from my bibliography as well as my academic/educational history. The primary focus of the thesis was to explore and investigate the intermittency in human movement. This section focuses on expressing my answer. A human being is constantly subject to the laws of physics, and works with them on a daily basis. In order for our CNS to execute a movement at a specified speed, it needs time and distance information. In this study, auditory feedback was dominant because it provided direct feedback on time, and haptic feedback was used less, due to the lack of experience with these movement distances and the increase in associated neurobiomechanical constraints. The metronome informed the participant of when to move and how long for, indicating that temporal frequency was a primary controlling variable. An analogy with associated information follows. The ability to dance requires the person to move across distances in time with the beat of the music. The brain executes high frequency tremors9 with muscles to detect changes in the environment across time. For example, miniature eye movements are executed at a high frequency and if an image were held stable (removing the effects of the miniature movements) on the retina the result would be a lack of visual information after a period. This time information collected by the CNS is used in all the senses, and is processed in order to perceive the world around us. Detecting a change in environment from an external an internal stimulus requires the perception of two points in time - before the stimulus (baseline) and during the stimulus (online) - and the ability to compare the two points. People can miss the change in 9
The tremors could be executed in the current experiment at 50 Hz (See Appendix B), higher then 50 Hz, one of the other frequencies noted or combination frequencies.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
103
environment (change blindness), and this blindness to change can be manipulated (Creighton, 2011). Following the aforementioned analogy, the dancer needs to compare the moment the music starts to the moment before the music. These two points in time effect a change in temporal frequency of air vibrations detected by the eardrum. The change in environment is then processed (online and offline) as soon as the stimulus has been detected (likely at a variable threshold). A choice is then made based on past experiences to determine if behaviour or movement should be executed in response to the stimulus. In this case the dancer needs to decide to start dancing. The dancer may decide not to dance if in the ‘moment of choice’ they are questioning their ability to do it well in front of the audience and they think they do not have the experience yet. If the stimulus is completely novel then the brain will need to decide if the stimulus may pose a threat or enable a reward. To find out, the CNS needs to interact with the stimulus by executing a period of behaviour and movement, and wait for a reward (clapping or awe), a threat (jeering), or nothing (silence). After interacting with the stimulus the consequences follow and change the likelihood of the person executing the movement again. In other words, a feedback loop (from a combination of the six senses) is used to determine if the stimulus has practical utility and if the behaviour should be repeated, continued or stopped, from the moment the foetus can move. Parkinson’s disease (Sheriden, et al., 1987), ADHD (Toplak, et al., 2006), and some drugs (Frowein, 1981) interfere with the ability to perceive time (Grondin, 2010; Wittmann, 2009, 2013). These internal environments and their effects on the CNS’s ability to perceive time, alter aspects of the control of movement at various stages in movement. These aspects include: when to imitate a movement, how long to execute a movement, when to stop a movement and the degree of isometric tremor. The intermittent signal results from Parkinson’ disease, this study and other motor control literature (e.g. Vallbo & Wessberg, 1993) suggests that a number of signals are sent and these appear to be related to the different
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
104
stages of movement. This study is based on software that makes calculations based on the input from a tablet. In order for the software to calculate the kinematic variables, it captured position data at a frequency of 100 Hz. This information is based on data collected at 0.01-second time intervals for the period of the experiment. In order for the computer to determine that an object is stationary, at two points all vectors need to have the same value. In order for the computer to capture the bandwidths from the pen tip (finger), the computer had to process time with a shorter interval than the execution of the bandwidths by the finger. I am suggesting that the CNS uses temporal frequency processes to interpret changes of position of stimuli (Sittig, et al., 1987) in the CNS’s environment, scalar variables such as temperature (Chen, et al., 2009; Okamoto, et al., 2012), and the basis for the senses in general. For example with smell or olfaction, the molecular vibration theory would be congruent with this idea (Malcolm Dyson, 1938; Turin, 2002). In other words, as the dancer traverses the dance floor, her CNS is taking snap shots of the environment, comparing each snap shot, and using the difference to inform her of the time (and from this distance) between objects (such as other people, the edge of the stage and light locations). Ideally, as she dances, the CNS is focuses on all stimuli that can be utilized for her to avoid danger or gain reward. Why was duration not related to intermittency at 100% in figure 14 and 15, and where does that variability around the means come from? The answers will likely come from systematic variability related to individual perceptions of time, practice effects, and the neurobiomechanical properties of each person. The individual perception of time is based on the idea that participants will likely have their own homeostasis, due to stimuli that affect time perception (for example, caffeine, stress, tiredness, arousal and so on). Practice effects were found in this study and are suggested to conform with the invariant rate scaling model (Terzuolo & Viviani, 1980). The neurobiomechanical properties can come from many
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
105
sources, such as age (18-43) and size of the person (the fixed distance will have a different degree of motion for each person). Considering all these variables and others, the relationship between time and intermittency is still very close to the 100%. This was 95% for acceleration zero crossings and 78% for normalized jerk-cost and this difference between the two measures indicating normalized jerk-cost was detecting more systematic variability. One source of systematic variability that differentiated the two measures of intermittency was distance, which affected the normalized jerk-cost relationship with duration (this model is available). Based on these ideas, I state that the perception of Time and temporal frequency is the primary ‘sense’ before all others. I am very confident in this theory at present, but I am also looking forward to being proven wrong. This idea is generally inline with previous work on the perception of time (Grondin, 2010; Wittmann, 2009, 2013), and temporal processors in the brain (Fugelsang, Roser, Corballis, Gazzaniga, & Dunbar, 2005) and has similarities with the theory by O'Regan and Noë (2001). One difference between the research noted and my perspective is the emphasis on time above distance, as the primary source of information. Where as previous researchers (Fugelsang, et al., 2005; Grondin, 2010) appear to treat them as equal, which may have more practice utility for research. Possible critiques of my theory might be circularity (that this theory might requires this theory to prove itself) and parsimonious (that this theory is very intuitively simple once you separated out the different variables) indicating to me the likelihood of the idea already existing within the scientific literate, but I have yet to find it. Future research across multiple disciplines would be needed to prove this concept wrong and find a better theory, or provide strong evidence that this concept is valid. Conclusions The aim of the present study was to determine if people could consciously improve
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
106
the smoothness of their movements. This aim was split into two goals: to determine if participants can move continuously at a constant velocity under the different timing and distance constraints; to determine the influences of the factors duration, distance and practice on intermittency. The results for the first goal indicated that people could generally move continuously at a constant velocity under the different timing and distance constraints. Their ability to perform the task had systematic error. The systematic error suggested that the control of movements were influenced by haptic and auditory feedback, which interacted with neurobiomechanical constraints and experience. The general aim of the planning and execution of movements appears to be focused on achieving ideal duration, with distance enabling ideal duration in the longer line but not in the shorter. Auditory feedback appeared to be the primary source to guide movements. The increase in neurobiomechanical appeared to be distinctly different for distance by duration effects and distance by duration by practice effects. For distance by duration effects, the constraints increased as both variable decreased. If practice effects are included, then the relationship with duration was the opposite. The results for the second goal indicated that duration was the primary factor in determining the degree of intermittency, but some situations might be controlled by velocity (combining both distance and duration information). Together duration (from auditory feedback), distance (from haptic feedback) and speed, depending on condition, appeared to influence practice. The practice effects imply that people could use the available feedback to consciously improve the smoothness of their arm movements under these conditions. This in turn suggests the fluctuations come from cognitively penetrable sources and neurobiomechanical sources. The cognitively penetrable sources could be from feedback loops, submovements, an internal model or a combination. The neurobiomechanical sources could be from at least two intermittent signals and system noise. Combined with previous research the results tend to support a single control method that becomes less efficient as
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
107
movements are executed for shorter durations and shorter distances (or longer durations and longer distances), due to an increase in neurobiomechanical sources and biomechanical constraints. This study provides a foundation for methods that could be used to improve the performance of similar tasks, such as braille reading and surgery. Additional post-hoc findings are in, support for future studies to combine both normalized jerk-cost and acceleration zero-crossings to measure intermittency and three bandwidths measured with Fast Fourier Transform, two of these bandwidths 3-5 Hz and 7-10 Hz, are similar to bandwidths found with Parkinson’s disease. To finish where this report started with the comic strip: people’s behaviour can be unexpected when placed under constraints and will occasionally execute odd behaviour, such as move backward. This odd behaviour is systematic and can be used to infer changes in control issues. People can improve their ability to make and execute movements smoother, but the CNS has challenges from neurobiomechanical or biomechanical sources that it may not be able to overcome.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
108
Appendix A was provided for people who prefer visual representations of the analysis. These models are works in progress and designed to provide a framework for future research. Appendix A:
Task Performance Data
Ideal Performance: Distance (Raised line on surface) Driving Frequency Short (8 cm)
Long (16 cm)
1 Hz (.5 s*)
1 cm/s
2 cm/s
0.25 Hz (2 s*)
4 cm/s
8 cm/s
0.0625 Hz (8 s*)
16 cm/s
32 cm/s
(Metronome) High Medium Low
*Values in ideal duration per segment.
Neurobiomechanical constraints: Distance by Duration effects Distance Duration
8 cm
Low
8s
Medium
2s
High
16 cm
0.5 s
The arrows represent the direction of the suggested increase in neurobiomechanical constraints. Notes: that this is relative. The source for the data is from Tables 1 and 2.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
109
Neurobiomechanical constraints: Distance by Duration by Practice effects. Distance Duration
8 cm
Low
8s
Medium
2s
High
16 cm
0.5 s
The arrows represent the direction of the suggested increase in neurobiomechanical constraints. Notes: that this is relative. The source for the data is from Figures 4, 5, and 6.
The Interaction between the Duration by Distance effects on the Auditory (A) and Haptic (H) Feedback Loops. Distance Duration 8 cm Low
8s
Medium
2s
High
0.5 s
Arrows indicate suggested increase in influence from that form of feedback
16 cm
H
a
a
H
h
A
A
h
H
h
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
110
The Interaction between the Duration by Distance by Practice effects on the Auditory (A) and Haptic (H) Feedback Loops. Distance Duration 8 cm Low
8s
Medium
2s
High
0.5 s
A
16 cm h
H
A
h
a
H
a
H
H
Notes: Duration is generally more predominant across all conditions. Arrows indicate suggested increase in influence from that form of feedback.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS. Appendix B:
111
Raw Spectrum Analysis Data from MozAlyzeR
The first two trials are used to show different level of a 50 Hz peak at different powers. The last trial was provided to show that the 50 Hz peak does appear to happen in all trials, although it could be hidden by noise. The 50 Hz peaks appeared to occur mainly in the low frequency conditions. Each graph is from a different subject.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
112
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
113
References Abend, W., Bizzi, E., & Morasso, P. (1982). Human arm trajectory formation. Brain, 105(JUN), 331-348. doi: 10.1093/brain/105.2.331 Abrams, R. A., & Pratt, J. (1993). Rapid aimed limb movements: Differential effects of practice on component submovements. Journal of Motor Behavior, 25(4), 288-298. doi: 10.1080/00222895.1993.9941650 Adams, J. A. (1987). Historical review and appraisal of research on the learning, retention, and transfer of human motor-skills. Psychological Bulletin, 101(1), 41-74. doi: 10.1037/0033-2909.101.1.41 Anton, H., Bivens, I., & Davis, S. (2005). Integration Calculus: Early transcendentals (8th ed.). Hoboken, NJ: John Wiley & Sons. Atkeson, C. G., & Hollerbach, J. M. (1985). Kinematic features of unrestrained vertical arm movements. The Journal of Neuroscience, 5(9), 2318-2330. doi: 0270.6474/85/05092318$02.00/O Blake, D. T., Hsiao, S. S., & Johnson, K. O. (1997). Neural Coding Mechanisms in Tactile Pattern Recognition: The Relative Contributions of Slowly and Rapidly Adapting Mechanoreceptors to Perceived Roughness. The Journal of Neuroscience, 17(19), 7480-7489. Blake, D. T., Johnson, K. O., & Hsiao, S. S. (1997). Monkey cutaneous SAI and RA responses to raised and depressed scanned patterns: Effects of width, height, orientation, and a raised surround. Journal of Neurophysiology, 78(5), 2503-2517. Cascio, C. J., & Sathian, K. (2001). Temporal cues contribute to tactile perception of roughness. The Journal of Neuroscience, 21(14), 5289-5296. Celik, O., Gu, Q., Deng, Z., & O'Malley, M. K. (2009, October 11-15). Intermittency of slow arm movements increases in distal direction. Paper presented at the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), St. Louis, MO, USA. Chapman, C. E., Tremblay, F., Jiang, W., Belingard, L. c., & Meftah, E.-M. (2002). Central neural mechanisms contributing to the perception of tactile roughness. Behavioural Brain Research, 135(1–2), 225-233. doi: 10.1016/s0166-4328(02)00168-7 Chen, X., Shao, F., Barnes, C., Childs, T., & Henson, B. (2009). Exploring relationships between touch perception and surface physical properties. International Journal of Design, 3(2), 67-76. Colman, A. M. (Ed.) (2001) Oxford dictionary of psychology. New York: Oxford Universtiy Press. Conditt, M. A., & Mussa-Ivaldi, F. A. (1999). Central representation of time during motor learning. Proceedings of the National Academy of Sciences of the United States of America, 96(20), 11625-11630. doi: 10.1073/pnas.96.20.11625 Cooper, J. O., Heron, T. E., & Heward, W. L. (2007). Applied behavior analysis (2nd ed.). Upper Saddle River, N.J.: Pearson/Merrill-Prentice Hall. Credo. (2013). Credo online reference service. Retrieved 11th of Feburary, from Credo http://www.credoreference.com.ezproxy.auckland.ac.nz/home.do Creighton, A. S. (2011). Magic and nuroscience. Presentation. The University of Auckland. Crossman, E. R. F. W., & Goodeve, P. J. (1983). Feedback control of hand-movement and Fitts' law. The Quarterly Journal of Experimental Psychology Section A, 35(2), 251278. doi: 10.1080/14640748308402133 d'Avella, A., Portone, A., & Lacquaniti, F. (2011). Superposition and modulation of muscle synergies for reaching in response to a change in target location. Journal of Neurophysiology. doi: 10.1152/jn.00675.2010
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
114
Darian-Smith, I., & Oke, L. E. (1980). Peripheral neural representation of the spatial frequency of a grating moving across the monkey's finger pad. The Journal of Physiology, 309(1), 117-133. Darling, W. G., Cole, K. J., & Abbs, J. H. (1988). Kinematic variability of grasp movements as a function of practice and movement speed. Experimental brain research, 73(2), 225-235. doi: 10.1007/bf00248215 DeLong, M. R. (1972). Activity of basal ganglia neurons during movement. Brain Research, 40(1), 127-135. doi: 10.1016/0006-8993(72)90118-7 DeLong, M. R. (1990). Primate models of movement disorders of basal ganglia origin. Trends in Neurosciences, 13(7), 281-285. doi: 10.1016/0166-2236(90)90110-v Dépeault, A., Meftah, E.-M., & Chapman, C. E. (2008). Tactile speed scaling: Contributions of time and space. Journal of Neurophysiology, 99(3), 1422-1434. doi: 10.1152/jn.01209.2007 Dipietro, L., Krebs, H. I., Fasoli, S. E., Volpe, B. T., & Hogan, N. (2009). Submovement changes characterize generalization of motor recovery after stroke. Cortex: A Journal Devoted to the Study of the Nervous System and Behavior, 45(3), 318-324. doi: http://dx.doi.org/10.1016/j.cortex.2008.02.008 Doeringer, J. A., & Hogan, N. (1995). Performance of above elbow body-powered prostheses in visually guided unconstrained motion tasks. IEEE Transactions on Biomedical Engineering, 42(6), 621-631. doi: 10.1109/10.387202 Doeringer, J. A., & Hogan, N. (1998). Intermittency in preplanned elbow movements persists in the absence of visual feedback. Journal of Neurophysiology, 80(4), 1787-1799. Dounskaia, N., Wisleder, D., & Johnson, T. (2005). Influence of biomechanical factors on substructure of pointing movements. Experimental brain research, 164(4), 505-516. doi: 10.1007/s00221-005-2271-4 Ekman, G., Hosman, J., & Lindstrom, B. (1965). Roughness, smoothness, and preference: A study of quantitative relations in individual subjects. Journal of Experimental Psychology, 70(1), 18-26. doi: 10.1037/h0021985 Elliott, D. (1990). Intermittent visual pickup and goal directed movement: A review. Human Movement Science, 9(3-5), 531-548. doi: http://dx.doi.org/10.1016/01679457%2890%2990013-4 Elliott, D., Binsted, G., & Heath, M. (1999). The control of goal-directed limb movements: Correcting errors in the trajectory. Human Movement Science, 18(2–3), 121-136. doi: 10.1016/s0167-9457(99)00004-4 Elliott, D., Helsen, W. F., & Chua, R. (2001). A century later: Woodworth's (1899) twocomponent model of goal-directed aiming. Psychological Bulletin, 127(3), 342-357. doi: 10.1037/0033-2909.127.3.342 Elliott, D., Lyons, J., & Dyson, K. (1997). Rescaling an acquired discrete aiming movement: Specific or general motor learning? Human Movement Science, 16(1), 81-96. doi: 10.1016/s0167-9457(96)00041-3 Engelbrecht, S. E. (2001). Minimum principles in motor control. Journal of Mathematical Psychology, 45(3), 497-542. doi: 10.1006/jmps.2000.1295 Findley, L. J. (1988). Tremors: Differential diagnosis and pharmacology. In J. J & T. E (Eds.), Parkinson’s disease and movement disorders (pp. 243-262). Baltimore, MD: Urban & Schwarzenberg. Fishbach, A., Roy, S. A., Bastianen, C., Miller, L. E., & Houk, J. C. (2007). Deciding when and how to correct a movement: discrete submovements as a decision making process. Experimental brain research, 177(1), 45-63. doi: 10.1007/s00221-006-0652y Fitts, P. M. (1954). The information capacity of the human motor system in controlling the
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
115
amplitude of movement. Journal of Experimental Psychology, 47(6), 381-391. doi: 10.1037/h0055392 Flash, T., & Henis, E. (1991). Arm trajectory modifications during reaching towards visual targets. Journal of cognitive Neuroscience, 3(3), 220-230. doi: 10.1162/jocn.1991.3.3.220 Flash, T., & Hochner, B. (2005). Motor primitives in vertebrates and invertebrates. Current Opinion in Neurobiology, 15(6), 660-666. doi: 10.1016/j.conb.2005.10.011 Flash, T., & Hogan, N. (1985). The coordination of arm movements: An experimentally confirmed mathematical model. Journal of Neuroscience, 5(7), 1688-1703. Fleckner, R. (2010). Metronome: For practicing musicians (Version 1.8.3). Retrieved from http://members.ozemail.com.au/~ronfleckner/metronome/ Fradet, L., Lee, G., & Dounskaia, N. (2008a). Origins of submovements during pointing movements. Acta Psychologica, 129(1), 91-100. doi: http://dx.doi.org/10.1016/j.actpsy.2008.04.009 Fradet, L., Lee, G., & Dounskaia, N. (2008b). Origins of submovements in movements of elderly adults. Journal of NeuroEngineering and Rehabilitation, 5(28), 1-14. doi: 10.1186/1743-0003-5-28 Freeman, J. B., Dale, R., & Farmer, T. A. (2011). Hand in motion reveals mind in motion. Frontiers in Psychology, 2(59), 1-6. doi: 10.3389/fpsyg.2011.00059 Frowein, H. W. (1981). Selective effects of barbiturate and amphetamine on information processing and response execution. Acta Psychologica, 47(2), 105-115. doi: http://dx.doi.org/10.1016/0001-6918(81)90002-0 Fugelsang, J. A., Roser, M. E., Corballis, P. M., Gazzaniga, M. S., & Dunbar, K. N. (2005). Brain mechanisms underlying perceptual causality. Cognitive Brain Research, 24(1), 41-47. doi: http://dx.doi.org/10.1016/j.cogbrainres.2004.12.001 Gamzu, E., & Ahissar, E. (2001). Importance of temporal cues for tactile spatial–frequency discrimination. The Journal of Neuroscience, 21(18), 7416-7427. Gawthrop, P., Loram, I., Lakie, M., & Gollee, H. (2011). Intermittent control: A computational theory of human control. Biological Cybernetics, 104(1), 31-51. doi: 10.1007/s00422-010-0416-4 Gentner, D. R. (1987). Timing of skilled motor performance: Tests of the proportional duration model. Psychological Review, 94(2), 255-276. doi: 10.1037/0033295x.94.2.255 Ghez, C., Krakauer, J. W., Sainburg, R. L., & Ghilardi, M. (2000). Spacial representations and internal models of limb dynamics in motor learning. In M. S. Gazzaniga (Ed.), The new cognitive neurosciences (2nd ed ) (pp. 501-514). Cambridge, MA: The MIT Press; US. Giszter, S. F. (2009). Motor primitives. In R. S. Editor-in-Chief: Larry (Ed.), Encyclopedia of Neuroscience (pp. 1023-1040). Oxford: Academic Press. Retrieved from http://www.sciencedirect.com/science/article/pii/B9780080450469013401 doi:10.1016/b978-008045046-9.01340-1 Grondin, S. (2010). Timing and time perception: A review of recent behavioral and neuroscience findings and theoretical directions. Attention, Perception, & Psychophysics, 72(3), 561-582. doi: 10.3758/app.72.3.561 Groß, J., Timmermann, L., Kujala, J., Dirks, M., Schmitz, F., Salmelin, R., et al. (2002). The neural basis of intermittent motor control in humans. Proceedings of the National Academy of Sciences, 99(4), 2299-2302. doi: 10.1073/pnas.032682099 Halsband, U., & Lange, R. K. (2006). Motor learning in man: A review of functional and clinical studies. Journal of Physiology-Paris, 99(4–6), 414-424. doi: 10.1016/j.jphysparis.2006.03.007
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
116
Helmich, R. C., Hallett, M., Deuschl, G., Toni, I., & Bastiaan, R. B. (2012). Cerebral causes and consequences of parkinsonian resting tremor: A tale of two circuits? Brain, 135(11), 1 - 21. doi: doi:10.1093/brain/aws023 Henis, E. (1991). Strategies underlying arm trajectory modification during reaching toward visual targets. Ph.D. DP17556, The Weizmann Institute of Science (Israel), Israel. Retrieved from http://ezproxy.auckland.ac.nz/login?url=http://proquest.umi.com/pqdweb?did=18868 49491&Fmt=7&clientId=13395&RQT=309&VName=PQD Hogan, N. (1984). An organizing principle for a class of voluntary movements. The Journal of Neuroscience, 4(11), 2745-2754. Hogan, N., & Sternad, D. (2009). Sensitivity of smoothness measures to movement duration, amplitude, and arrests. Journal of Motor Behavior, 41(6), 529-534. doi: 10.3200/3509-004-rc Hollins, M., & Bensmaïa, S. J. (2007). The coding of roughness. Canadian Journal of Experimental Psychology/Revue canadienne de psychologie expérimentale, 61(3), 184-195. doi: 10.1037/cjep2007020 Hollins, M., & Risner, S. (2000). Evidence for the duplex theory of tactile texture perception. Attention, Perception, & Psychophysics, 62(4), 695-705. doi: 10.3758/bf03206916 Hourcade, J. P. (2006). Learning from preschool children's pointing sub-movements. Paper presented at the Proceedings of the 2006 on interaction design and children, Tampere, Finland. Hughes, B. (2011). Movement kinematics of the braille-reading finger. Journal of Visual Impairment & Blindness, 105(6), 370-381. Hughes, B., & Jansson, G. (1994). Texture perception via active touch. Human Movement Science, 13(3–4), 301-333. doi: http://dx.doi.org/10.1016/0167-9457(94)90044-2 Hughes, B., Van Gemmert, A. W. A., & Stelmach, G. E. (2011). Linguistic and perceptualmotor contributions to the kinematic properties of the braille reading finger. Human Movement Science, 30, 711-730. doi: 10.1016/j.humov.2010.05.005 Hwang, F., Keates, S., Langdon, P., & Clarkson, J. (2005). A submovement analysis of cursor trajectories. Behaviour & Information Technology, 24(3), 205-217. doi: http://dx.doi.org/10.1080/01449290412331327474 Ingle, V. K., & Proakis, J. G. (2012). Digital signal processing using MATLAB (3rd ed.). Stanford CT: Cengage Learning. Ivry, R. B., Keele, S. W., & Diener, H. C. (1988). Dissociation of the lateral and medial cerebellum in movement timing and movement execution. Experimental brain research, 73(1), 167-180. doi: 10.1007/bf00279670 Johnson, K. O., & Hsiao, S. S. (1992). Neural Mechanism of tactual form and textureperception. Annual Review of Neuroscience, 15, 227-250. doi: 10.1146/annurev.neuro.15.1.227 Kakuda, N., Nagaoka, M., & Wessberg, J. (1999). Common modulation of motor unit pairs during slow wrist movement in man. The Journal of Physiology, 520(3), 929-940. doi: 10.1111/j.1469-7793.1999.00929.x Kawato, M. (1999). Internal models for motor control and trajectory planning. Current Opinion in Neurobiology, 9(6), 718-727. doi: 10.1016/s0959-4388(99)00028-8 Khan, M. A., & Franks, I. M. (2000). The effect of practice on component submovements is dependent on the availability of visual feedback. Journal of Motor Behavior, 32(3), 227-240. doi: 10.1080/00222890009601374 Khan, M. A., & Franks, I. M. (2003). Online versus offline processing of visual feedback in the production of component submovements. Journal of Motor Behavior, 35(3), 285295. doi: http://dx.doi.org/10.1080/00222890309602141
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
117
Khan, M. A., Franks, I. M., & Goodman, D. (1998). The effect of practice on the control of rapid aiming movements: Evidence for an interdependency between programming and feedback processing. The Quarterly Journal of Experimental Psychology Section A, 51(2), 425-443. doi: 10.1080/713755756 Klatzky, R., & Lederman, S. J. (1999). Tactile roughness perception with a rigid link interposed between skin and surface. Attention, Perception, & Psychophysics, 61(4), 591-607. doi: 10.3758/bf03205532 Lamb, G. D. (1983). Tactile discrimination of textured surfaces: Psychophysical performance measurements in humans. The Journal of Physiology, 338(1), 551-565. Lederman, S. J. (1974). Tactile roughness of grooved surfaces: The touching process and effects of macro- and microsurface structure. Attention, Perception, & Psychophysics, 16(2), 385-395. doi: 10.3758/bf03203958 Lederman, S. J. (1981). The perception of surface roughness by active and passive touch. Bulletin of the Psychonomic Society, 18(5), 253-255. Lederman, S. J. (1983). Tactual roughness perception: Spatial and temporal determinants. Canadian Journal of Psychology/Revue canadienne de psychologie, 37(4), 498-511. doi: 10.1037/h0080750 Lederman, S. J., Howe, R. D., Klatzky, R. L., & Hamilton, C. (2004, 27-28 March 2004). Force variability during surface contact with bare finger or rigid probe. Paper presented at the Haptic Interfaces for Virtual Environment and Teleoperator Systems, 2004. HAPTICS '04. Proceedings. 12th International Symposium on. Lederman, S. J., & Taylor, M. (1972). Fingertip force, surface geometry, and the perception of roughness by active touch. Attention, Perception, & Psychophysics, 12(5), 401408. doi: 10.3758/bf03205850 Libouton, X., Barbier, O., Berger, Y., Plaghki, L., & Thonnard, J.-L. (2012). Tactile roughness discrimination of the finger pad relies primarily on vibration sensitive afferents not necessarily located in the hand. Behavioural Brain Research, 229(1), 273-279. doi: 10.1016/j.bbr.2012.01.018 Loomis, J. M. (1985). Tactile recognition of raised characters: A parametric study. Bulletin of the Psychonomic Society, 23(1), 18-20. Loram, I. D., Gollee, H., Lakie, M., & Gawthrop, P. J. (2011). Human control of an inverted pendulum: Is continuous control necessary? Is intermittent control effective? Is intermittent control physiological? The Journal of Physiology, 589(2), 307-324. doi: http://dx.doi.org/10.1113/jphysiol.2010.194712 Ma, H.-i., & Trombly, C. A. (2004). Effects of task complexity on reaction time and movement kinematics in elderly people. American Journal of Occupational Therapy, 58(2), 150-158. Malcolm Dyson, G. (1938). The scientific basis of odour. Journal of the Society of Chemical Industry, 57(28), 647-651. doi: 10.1002/jctb.5000572802 Mathur, A., & Hughes, B. (2009). The relationship between mean movement velocity and its intermittency in a haptic task. Unpublished data, University of Auckland. McAuley, J. H., Rothwell, J. C., & Marsden, C. D. (1999). Human anticipatory eye movements may reflect rhythmic central nervous activity. Neuroscience, 94(2), 339350. doi: http://dx.doi.org/10.1016/S0306-4522(99)00337-1 Meftah, E. M., Belingard, L., & Chapman, C. E. (2000). Relative effects of the spatial and temporal characteristics of scanned surfaces on human perception of tactile roughness using passive touch. Experimental brain research, 132(3), 351-361. doi: 10.1007/s002210000348 Meyer, D. E., Abrams, R. A., Kornblum, S., Wright, C. E., & Smith, K. J. E. (1988). Optimality in human motor performance: Ideal control of rapid aimed movements.
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
118
Psychological Review, 95(3), 340-370. doi: http://dx.doi.org/10.1037/0033295X.95.3.340 Meyer, D. E., Smith, K. J. E., Kornblum, S., Abrams, R. A., & Wright, C. E. (1990). Speedaccuracy tradeoffs in aimed movements: Toward a theory of rapid voluntary action Attention and performance 13: Motor representation and control (pp. 173-226). Hillsdale, NJ, England: Lawrence Erlbaum Associates, Inc; England. Miall, R. C., Weir, D. J., & Stein, J. F. (1993). Intermittency in human manual tracking tasks. Journal of Motor Behavior, 25(1), 53-63. doi: 10.1080/00222895.1993.9941639 Miall, R. C., & Wolpert, D. M. (1996). Forward models for physiological motor control. Neural Networks, 9(8), 1265-1279. doi: 10.1016/s0893-6080(96)00035-4 Milner, T. E. (1992). A model for the generation of movements requiring endpoint precision. Neuroscience, 49(2), 487-496. doi: 10.1016/0306-4522(92)90113-G Milner, T. E., & Ijaz, M. M. (1990). The effect of accuracy constraints on three-dimensional movement kinematics. Neuroscience, 35(2), 365-374. doi: 10.1016/03064522(90)90090-q Morgan, M., Phillips, J. G., Bradshaw, J. L., Mattingley, J. B., Iansek, R., & Bradshaw, J. A. (1994). Age-related motor slowness: Simply strategic? Journal of Gerontology, 49(3), M133-M139. doi: 10.1093/geronj/49.3.M133 Morley, J., & Goodwin, A. (1987). Sinusoidal movement of a grating across the monkey's fingerpad: temporal patterns of afferent fiber responses. The Journal of Neuroscience, 7(7), 2181-2191. Mussa-Ivaldi, F. A. (1999). Modular features of motor control and learning. Current Opinion in Neurobiology, 9(6), 713-717. doi: 10.1016/s0959-4388(99)00029-x Mussa–Ivaldi, F. A., & Bizzi, E. (2000). Motor learning through the combination of primitives. Philosophical Transactions of the Royal Society of London. Series B: Biological Sciences, 355(1404), 1755-1769. doi: 10.1098/rstb.2000.0733 Nagasaki, H. (1991). Asymmetrical trajectory formation in cyclic forearm movements in man. Experimental brain research, 87(3), 653-661. doi: 10.1007/bf00227091 Navas, F., & Stark, L. (1968). Sampling or intermittency in hand control system dynamics. Biophysical Journal, 8(2), 252-302. doi: 10.1016/s0006-3495(68)86488-4 NeuroScript. (2006-2010). MovAlyzeR: Software for handwriting movement recording and analysis (Version 5.8 & 6.1). Retrieved from http://www.neuroscriptsoftware.com/ Newell, K. M. (1991). Motor skill acquisition. Annual Review of Psychology, 42, 213-237. doi: 10.1146/annurev.ps.42.020191.001241 Nishikawa, K. C., Murray, S. T., & Flanders, M. (1999). Do arm postures vary with the speed of reaching? Journal of Neurophysiology, 81(5), 2582-2586. O'Regan, J. K., & Noë, A. (2001). A sensorimotor account of vision and visual consciousness. Behavioral and Brain Sciences, 24(5), 939-1031. Okamoto, S., Nagano, H., & Yamada, Y. (2012). Psychophysical dimensions of tactile perception of textures. Haptics, IEEE Transactions on, PP(99), 1-13. doi: 10.1109/toh.2012.32 Pasalar, S., Roitman, A. V., & Ebner, T. J. (2005). Effects of speeds and force fields on submovements during circular manual tracking in humans. Experimental brain research, 163(2), 214-225. doi: 10.1007/s00221-004-2169-6 Pearson, K. G. (1972). Central programming and reflex control of walking in the cockroach. Journal of Experimental Biology, 56(1), 173-193. Pratt, J., & Abrams, R. A. (1996). Practice and component submovements: The roles of programming and feedback in rapid aimed limb movements. Journal of Motor Behavior, 28(2), 149-156. doi: 10.1080/00222895.1996.9941741 Pratt, J., Chasteen, A. L., & Abrams, R. A. (1994). Rapid aimed limb movements: Age
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
119
differences and practice effects in component submovements. Psychology and Aging, 9(2), 325-334. doi: 10.1037/0882-7974.9.2.325 Proteau, L., & Cournoyer, J. (1990). Vision of the stylus in a manual aiming task: The effects of practice. The Quarterly Journal of Experimental Psychology Section A, 42(4), 811828. doi: 10.1080/14640749008401251 Proteau, L., & Marteniuk, R. G. (1993). Static visual information and the learning and control of a manual aiming movement. Human Movement Science, 12(5), 515-536. doi: http://dx.doi.org/10.1016/0167-9457(93)90003-8 Proteau, L., Marteniuk, R. G., Girouard, Y., & Dugas, C. (1987). On the type of information used to control and learn an aiming movement after moderate and extensive training. Human Movement Science, 6(2), 181-199. doi: http://dx.doi.org/10.1016/01679457(87)90011-X Reber, A. S., & Reber, E. (Eds.). (2001). London: Penguin Group. Rohrer, B., Fasoli, S., Krebs, H., Volpe, B., Stein, J., & Hogan, N. (2003). Submovement overlap as a measure of movement smoothness. IGS. Retrieved from http://www.sandia.gov/~brrohre/papers.html Rohrer, B., Fasoli, S., Krebs, H. I., Hughes, R., Volpe, B., Frontera, W. R., et al. (2002). Movement smoothness changes during stroke recovery. The Journal of Neuroscience, 22(18), 8297-8304. Schneider, K., & Zernicke, R. F. (1989). Jerk-cost modulations during the practice of rapid arm movements. Biological Cybernetics, 60(3), 221-230. doi: 10.1007/bf00207290 Shadmehr, R., & Mussa-Ivaldi, F. (1994). Adaptive representation of dynamics during learning of a motor task. The Journal of Neuroscience, 14(5, Pt 2), 3208-3224. Shannon, C. E. (2001). A mathematical theory of communication. SIGMOBILE Mob. Comput. Commun. Rev., 5(1), 3-55. doi: 10.1145/584091.584093 Sheriden, M. R., Flowers, K. A., & Hurrell, J. (1987). Programming and execution of movement in Parkinson's disease. Brain, 110(5), 1247-1271. doi: 10.1093/brain/110.5.1247 Sinclair, R. J., & Burton, H. (1991). Tactile Discrimination of Gratings: Psychophysical and Neural Correlates in Human and Monkey. Somatosensory & Motor Research, 8(3), 241-248. doi: doi:10.3109/08990229109144747 Sinclair, R. J., Pruett, J. R., & Burton, H. (1996). Responses in Primary Somatosensory Cortex of Rhesus Monkey to Controlled Application of Embossed Grating and Bar Patterns. Somatosensory & Motor Research, 13(3-4), 287-306. doi: doi:10.3109/08990229609052584 Sittig, A. C., Denier, J. J., & Gielen, C. C. A. M. (1987). The contribution of afferent information on position and velocity to the control of slow and fast human forearm movements. Experimental brain research, 67(1), 33-40. doi: 10.1007/bf00269450 Slifkin, A. B., & Newell, K. M. (1999). Noise, information transmission, and force variability. Journal of Experimental Psychology: Human Perception and Performance, 25(3), 837-851. doi: http://dx.doi.org/10.1037/0096-1523.25.3.837 Slifkin, A. B., Vaillancourt, D. E., & Newell, K. M. (2000). Intermittency in the control of continuous force production. Journal of Neurophysiology, 84(4), 1708-1718. Smith, A. M., Chapman, C. E., Deslandes, M., Langlais, J.-S., & Thibodeau, M.-P. (2002). Role of friction and tangential force variation in the subjective scaling of tactile roughness. Experimental brain research, 144(2), 211-223. doi: 10.1007/s00221-0021015-y Smith, A. M., & Scott, S. H. (1996). Subjective scaling of smooth surface friction. Journal of Neurophysiology, 75(5), 1957-1962. Srinivasan, M. A., Whitehouse, J. M., & LaMotte, R. H. (1990). Tactile detection of slip:
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
120
surface microgeometry and peripheral neural codes. Journal of Neurophysiology, 63(6), 1323-1332. Terzuolo, C. A., & Viviani, P. (1980). Determinants and characteristics of motor patterns used for typing. Neuroscience, 5(6), 1085-1103. doi: http://dx.doi.org/10.1016/03064522(80)90188-8 Teulings, H.-L., Contreras-Vidal, J. L., Stelmach, G. E., & Adler, C. H. (1997). Parkinsonism reduces coordination of fingers, wrist, and arm in fine motor control. Experimental Neurology, 146(1), 159-170. doi: 10.1006/exnr.1997.6507 Thoroughman, K. A., & Shadmehr, R. (2000). Learning of action through adaptive combination of motor primitives. Nature, 407(6805), 742-747. doi: http://dx.doi.org/10.1038/35037588 Toplak, M. E., Dockstader, C., & Tannock, R. (2006). Temporal information processing in ADHD: Findings to date and new methods. Journal of Neuroscience Methods, 151(1), 15-29. doi: http://dx.doi.org/10.1016/j.jneumeth.2005.09.018 Tunik, E., Houk, J. C., & Grafton, S. T. (2009). Basal ganglia contribution to the initiation of corrective submovements. NeuroImage, 47(4), 1757-1766. doi: http://dx.doi.org/10.1016/j.neuroimage.2009.04.077 Turin, L. (2002). A Method for the Calculation of Odor Character from Molecular Structure. Journal of Theoretical Biology, 216(3), 367-385. doi: http://dx.doi.org/10.1006/jtbi.2001.2504 Vallbo, A., & Wessberg, J. (1993). Organization of motor output in slow finger movements in man. The Journal of Physiology, 469(1), 673-691. van der Wel, R. P., Sternad, D., & Rosenbaum, D. A. (2010). Moving the arm at different rates: Slow movements are avoided. Journal of Motor Behavior, 42(1), 29-36. doi: http://dx.doi.org/10.1080/00222890903267116 Viviani, P., & Terzuolo, C. (1980). Space-time invariance in learned motor skills. In G. E. Stelmach & J. Requin (Eds.), Advances in Psychology (Vol. Volume 1, pp. 525-533): North-Holland. Retrieved from http://www.sciencedirect.com/science/article/pii/S0166411508619676 doi:http://dx.doi.org/10.1016/S0166-4115(08)61967-6 Viviani, P., & Terzuolo, C. (1982). Trajectory determines movement dynamics. Neuroscience, 7(2), 431-437. doi: http://dx.doi.org/10.1016/0306-4522(82)90277-9 von Hofsten, C. (1991). Structuring of early reaching movements: A longitudinal study. Journal of Motor Behavior, 23(4), 280-292. doi: 10.1080/00222895.1991.9942039 von Hofsten, C., & Rönnqvist, L. (1993). The structuring of neonatal arm movements. Child Development, 64(4), 1046-1057. doi: 10.2307/1131326 Wheat, H. E., Salo, L. M., & Goodwin, A. W. (2010). Cutaneous afferents from the monkeys fingers: Responses to tangential and normal forces. Journal of Neurophysiology, 103(2), 950-961. doi: 10.1152/jn.00502.2009 Wiegner, A. W., & Wierzbicka, M. M. (1992). Kinematic models and human elbow flexion movements: Quantitative analysis. Experimental brain research, 88(3), 665-673. Wisleder, D., & Dounskaia, N. (2007). The role of different submovement types during pointing to a target. Experimental brain research, 176(1), 132-149. doi: 10.1007/s00221-006-0603-7 Wittmann, M. (2009). The inner experience of time. Philosophical Transactions of the Royal Society B: Biological Sciences, 364(1525), 1955-1967. doi: 10.1098/rstb.2009.0003 Wittmann, M. (2013). The inner sense of time: how the brain creates a representation of duration. [10.1038/nrn3452]. Nat Rev Neurosci, 14(3), 217-223. Wolpert, D. M., Ghahramani, Z., & Flanagan, J. R. (2001). Perspectives and problems in motor learning. Trends in Cognitive Sciences, 5(11), 487-494. doi: 10.1016/s1364-
PRACTICING MOVEMENT INTERMITTENCY WITH FEEDBACK LOOPS.
121
6613(00)01773-3 Woodworth, R. S. (1899). Accuracy of voluntary movement. The Psychological Review: Monograph Supplements, 3(3), i-114. doi: 10.1037/h0092992 Yoshioka, T., Craig, J. C., Beck, G. C., & Hsiao, S. S. (2011). Perceptual constancy of texture roughness in the tactile system. The Journal of Neuroscience, 31(48), 1760317611. doi: 10.1523/jneurosci.3907-11.2011 Yoshioka, T., Gibb, B., Dorsch, A. K., Hsiao, S. S., & Johnson, K. O. (2001). Neural coding mechanisms underlying perceived roughness of finely textured surfaces. The Journal of Neuroscience, 21(17), 6905-6916. Zollo, L., Salerno, A., Rossini, L., & Guglielmelli, E. (2010, 26-29 Sept. 2010). Submovement composition for motion and interaction control of a robot manipulator. Paper presented at the Biomedical Robotics and Biomechatronics (BioRob), 2010 3rd IEEE RAS and EMBS International Conference on.
