Efficacy and feeling of a vibrotactile Frontal Collision Warning implemented in a haptic pedal Helios de Rosarioa,∗ , Marcos Louredob,c , I˜naki D´ıazb , Andr´es Solera , Jorge Juan Gilb,c , Jos´e S. Solaza , Jordi Jornetd a Instituto

de Biomec´anica de Valencia, Universidad Polit´ecnica de Valencia. Edificio 9C, Camino de Vera s/n. E-46022 Valencia, Spain b CEIT. Paseo Manuel Lardiz´ abal, 15. E-20018 San Sebasti´an, Spain c TECNUN, University of Navarra. Paseo Manuel Lardiz´ abal, 13. E-20018 San Sebasti´an, Spain d FICOSA INTERNATIONAL S.A. Pol. Industrial Can Magarola, Ctra. C-17 km 13. E-08100 Mollet del Vall` es, Barcelona, Spain

Abstract A haptic pedal has been designed to emulate the behaviour of a common vehicle pedal and render superimposed vibrations with different characteristics. It was installed in a driving simulator, as an accelerator pedal with the secondary function of a vibrotactile Frontal Collision Warning (FCW). The efficacy and feeling of this solution was tested with 30 subjects using vibrotactile signals with 0.50, 1.05, and 1.60 Nm, at 2.5, 5, and 10 Hz, against a baseline visual FCW. Participants had to match the speed of a leading vehicle when the FCW was triggered. Their braking response was evaluated in terms of brake reaction time, matching speed time, control of velocity and headway reduction. Driver’s feelings were assessed with Kansei methodologies. Haptic stimuli were found to be more effective than visual signals, and the characteristics of the vibration also influenced the results. The best performance was achieved at the maximum amplitude, and in the range between 5 and 10 Hz. The perceived functionality and discomfort followed a trend coherent with the objective measurements. The conclusions of this study may be applied to develop effective and safe warning systems in vehicles, limiting the annoyance that they might cause to drivers. Key words: FCW, vibrotactile, semantic, haptic devices

1. Introduction The improvement of safety and pleasure on the road are driving forces of technological advances in the automotive industry. In particular, new Advanced Driver Assistance Systems (ADAS) have a great potential for making the experience of driving more relaxed and safer, by means of mitigating human errors (Amditis et al., 2007). One of the critical elements of ADAS is the communication with the user, which has to be clear and efficient, but at the same time ∗ Corresponding

author Email addresses: [email protected] (Helios de Rosario), [email protected] (Marcos Louredo), [email protected] (I˜naki D´ıaz), [email protected] (Andr´es Soler), [email protected] (Jorge Juan Gil), [email protected] (Jos´e S. Solaz), [email protected] (Jordi Jornet) Preprint submitted to Transportation Research Part F: Traffic Psychology and Behaviour September 21, 2009

must not overload the driver’s attention and perception resources. In fact, the second and third design goals of the European Statement of Principles on the Design of HMI (ESoP) indicate that the allocation of driver attention while interacting with displays and controls must be compatible with the attentional demand of the driving situation, and specifically that the system must not distract or visually entertain the driver (2007/78/EC, 2007). In order to lessen the visual load, many concepts of ADAS use auditory or haptic displays. Haptic interfaces, which transmit information through the muscular and tactile sensory systems, have proven to be very efficient for high performance human-machine interaction, exploring the capabilities of the human sense of touch as an advanced communication channel. Over the past years haptic interfaces have been successfully integrated into a wide range of fields such as engineering (Borro et al., 2004), surgery (Madhani et al., 1998; Li and Liu, 2006), and vehicle simulators (Frisoli et al., 2001; Tideman et al., 2004). One feature of haptic systems is that the part of the body that receives the information may be the same one that manipulates the interface, and thus action and feedback can be coupled. This advantage has been applied to manual interaction in the steering wheel, through active steering systems that automatically modify the wheel angle or torque resistance, in order to attenuate yaw disturbances, or for “shared control” between vehicle and driver in path following tasks (Ackermann et al., 1999; Steele and Gillespie, 2001). A similar approach for the lower limb has led to the development of active haptic pedals, which exert a variable counterforce depending on vehicle dynamics or surrounding traffic, in order to manage congestions (Brookhuis et al., 2008; van Driel et al., 2007) or control speed (Adell and V´arhelyi, 2008; Adell et al., 2008; Hj¨almdahl and V´arhelyi, 2004). However, while continuous haptic gas-pedal feedback is effective for car-following, it can be insufficient in more dangerous situations, when the distance to the leading vehicle is small, and quick corrective control actions must be taken to prevent collision (Mulder et al., 2008). Warning signals in the form of tactile vibrations or pulses have been largely tested in pedals (Martens and van Winsum, 2001; Tijerina et al., 2000; Lloyd et al., 1999), steering wheel (Tijerina et al., 2000; Jordan et al., 2007; Suzuki and Jansson, 2003), and driver seat (Lee et al., 2004; Stanley, 2006; Jordan et al., 2007). The part to which the vibration is associated normally depends on the type of warning and the expected action. It is exerted on pedals to compel braking events, in speed or collision warnings, and on the steering wheel for warnings related to lateral events that need steering actions, like a risk of lane departure. Seat vibration is used for both types of warnings, and also for other purposes, like providing orientational cues in a navigation system (van Erp and van Veen, 2004). This type of directional haptic feedback may be also applied to other functions, like calling the attention to the windscreen or the rear mirror, by a vibrating belt with tactors on the driver’s abdomen and back (Ho et al., 2005). Those studies show that vibrotactile warnings systematically improve reaction times and driving performance when compared to control cases without warnings (Jordan et al., 2007; Martens and van Winsum, 2001), or when compared with visual warnings (Ho et al., 2005; van Erp and van Veen, 2004). The advantage of vibrotactile over auditory warnings depends on the context: some studies find similar objective results for both kind of modalities (Lee et al., 2004) or combinations of them (Jordan et al., 2007); and there is no consensus on user preferences, which can be favorable to auditory systems (Adell et al., 2008), to vibrotactile solutions (Hoffman et al., 2003; Lee et al., 2004; Stanley, 2006), or neutral (Martens and van Winsum, 2001), depending on the study. Speech warnings provide more comprehensive information, and thus may be better suited for situations related to law-enforcement, like speed limit violations (Martens and van Winsum, 2001). On the other hand, vibrations are more effective than tones for correcting actions when the meaning of the warning has not been learnt beforehand (Suzuki and Jansson, 2003), and in 2

time-critical events, like an unexpected braking of the leading vehicle (Martens and van Winsum, 2001). To achieve the maximum efficacy, haptic signals must have a high stimulus-response compatibility with the action to be performed (Lloyd et al., 1999; Tijerina et al., 2000). This implies that the stimulus must be felt by the part of the body that must react (the foot in the case of pedal actions), and the direction and profile of the haptic feedback should prime the motor reactions of drivers (Jordan et al., 2007; Suzuki and Jansson, 2003). There is usually a proportional relation between the intensity of the stimulus and the response (Lee et al., 2004; Lloyd et al., 1999; Tijerina et al., 2000). One recommendation drawn from informal road-tests for braking haptic warnings, is using multiple pulses 100 ms long, separated by 100 to 200 ms periods (Lloyd et al., 1999). This study presents the development and evaluation of a haptic pedal with a vibrotactile Frontal Collision Warning (FCW). It analyses the efficacy and sensations produced by this warning, which has a high potential to improve safety and reduce accident-related congestion, but must have a good acceptance to be really useful (Jamson et al., 2008). In previous studies, the characteristics of vibration (frequency, amplitude and duration) in a vibrotactile steering wheel have not been found to be determinant of the efficacy of a FCW, but the amplitude of a monopulse pedal did affect the braking response (Tijerina et al., 2000). In the present study we tested the efficacy of a FCW modeled as a periodic vibrotactile signal in the pedal, which varied not only in amplitude, but also in frequency. The objective was to look for a convenient configuration in both objective and subjective terms, within ranges based on detection thresholds and recommendations reported in the literature, including some “stronger” stimuli (as long as they did not clearly hinder the vehicle control). As already commented on, we could expect a positive relation between the intensity of the FCW and driver’s reaction. But haptic warnings can also have negative effects, like interference with driving actions or annoyance, so there may be one point where an adequate balance between the opposing forces can be achieved. Like in previous studies of this type of warnings, the efficacy of the FCW was assessed in terms of speed control, separation to the leading vehicle and braking times (Tijerina et al., 2000; Jamson et al., 2008; Abe and Richardson, 2006; Warshawsky-Livne and Shinar, 2002). Driver’s sensations were also evaluated, since the satisfaction of drivers is a strong impetus for the development of electronic driving aids, in addition to safety (Brookhuis and de Waard, 2004). In the cited literature subjective impressions were also measured, in terms of appropriateness (from “much too soft” to “much too hard”), timeliness of the warning, mental effort, user acceptance, trust, and personal factors like safety, irritation, stress, joy of driving, etc. The present study evaluated the sensations of users following an approach based on Kansei Engineering, often used in the automotive industry (Horiguchi and Suetomi, 1995; Nagamachi, 1995): from a large catalogue of adjectives related to pedals, warnings and emotions, including the elements mentioned above, a narrower set of principal concepts was defined and these concepts were the object of analysis. The results of the tests were analysed in order to determine how the efficacy and sensations caused by the FCW varied with the characteristics of the pedal vibration, and to find the optimal combination within the tested ranges of amplitude and frequency. 2. Methods 2.1. Apparatus A new haptic foot pedal (Figure 1) was conceived, designed and built by the Applied Mechanics Department of CEIT. It is a floor-grounded device providing one rotary active degree 3

® = 0º Foot support plate Foot support plate

Encoder

® ® = 13º

¿user Motor

Base of the pedal

¿motor

Base of the pedal Á

Cable transmission Horizontal plane Figure 1: Haptic device emulating gas pedal and scheme of the device in working position.

of freedom (DOF) in the direction of the dorsiflexion-plantarflexion movement of the ankle. It was designed to allow users to move freely in the desired workspace, emulating the behaviour of a common pedal, and additionally rendering configurable torques to the users. In order to provide the user with a realistic and reliable driving feedback, special effort was made on gravity counterbalance of the device, and to achieve a high stiffness-to-weight ratio with a simplified design. The only DOF is driven by a commercial Maxon RE 40 DC motor and a specially designed cable transmission, that allows to exert a maximum continuous torque of 2680 mNm and a maximum peak torque of 8040 mNm. Among the different types of mechanical transmissions, cable transmissions are commonly used in haptic devices, since they offer several advantages such as low friction and no backlash. The rotation of the device is measured by a Quantum Device QD145 encoder coupled to the motor. The device was designed for a working position in which the user is seated, allowing a range of movement of 13◦ (angle α in Figure 1). The maximum angle between the horizontal plane and the foot support plate can be modified through the angle φ between the horizontal plane and the base of the pedal, in order to simulate the working position of the gas pedal of any car (Figure 1). The torque τ exerted by the device to emulate the behavior of an accelerator pedal is calculated as: τ = Kα (1) where K is the configurable stiffness. Modifying this parameter allows to simulate the behavior of different pedals. In the experiments this stiffness was set to 0.53 Nm/◦ . The pedal was integrated in a fixed-base driving simulator of the Institute of Biomechanics of Valencia, which consisted of a frame with a driver seat, steering wheel, accelerator and braking pedals, the accelerator being the described haptic pedal (Figure 2). The angle φ between the horizontal plane and the base of the pedal was set to 39◦ , to match the inclination of the pedals of the simulator. The driving controls were connected to the STISIM DriveTM simulator, and the frame was placed with the steering wheel at 2.5 m from a 1.6 × 1.2 m screen, on which the simulated scene was projected. The simulator and the haptic pedals were controlled by two separate computers, although the signal of the pedal angle was also sent as an input to the 4

Figure 2: Driving simulator with the haptic pedal.

simulator, in order to control the acceleration of the vehicle. The variables of the simulation (a continuous record of the configuration of the controls, and the position of the vehicle) were recorded in the simulator computer, and the pedal signals in the haptic computer. These data were synchronized for the analysis described in Section 2.3. 2.2. Subjects and procedure Thirty Spanish licensed drivers (15 men and 15 women) ranging in age from 20 to 40 years participated in the study. Before the trial, all participants filled in a consent form, in which they were informed of the objective of the test, and were paid for their participation. The experiment consisted of a simulation of driving along a road at constant speed (120 km/h), with continuous traffic in the opposite direction, to prevent the tendency of drivers to pass other vehicles (BarGera and Shinar, 2005) that sometimes appeared at a lower speed (80 km/h). Participants were requested to keep the speed of 120 km/h until the FCW was triggered, because the leading vehicle was getting too close. At that moment, the participants had to decrease their speed to match the 80 km/h of the leading vehicle. The FCW was activated by an external supervisor, when the headway of the leading vehicle approached 50 m. Assuming that the participants drove at 120 km/h, this distance would account for a time headway (THW) of 1.5 s, and a time-to-collision (TTC) of 4.5 s, according to the expressions: h (2) THW = v h T TC = (3) (vo − vl ) with h representing the headway (distance between the vehicles), and vo , vl the speeds of the own vehicle and the leading vehicle, respectively. The FCW was also stopped by the supervisor when 5

drivers stabilized their speed around the target value (80 km/h). The activation headway was defined in prior pilot experiments, in order to achieve an actual risk of crash, that could be normally avoided with a fast reaction. The associated THW was equal to the reference used in other studies of car-following (Mulder et al., 2008), and the TTC was within or slightly over the threshold recommendations of 4 or 5 s for collision warnings (Hirst and Graham, 1997). It was a cautious measure, in order to keep the number of crashes to a minimum. The braking response and the subjective impression of the FCW were recorded, as described in the following sections. Before starting the measurements, participants practised during approximately 10 minutes, in order to familiarize themselves with the simulator and the exercise. Every participant tested 10 variants of the FCW. Nine of these variants were a vibrotactile FCW, combining three torque amplitudes and three frequencies. The tested torques were 0.50, 1.05, and 1.60 Nm, resulting from applying 0.02, 0.04, and 0.06 Nm to the motor, respectively. Those torques accounted for a range between approximately 7% and 25% (exactly between 7.2% and 23.2%) of the greatest torque exerted by the pedal, which was 6.9 Nm, for the pedal completely pressed, according to Equation (1) with α = 13◦ . The vibration wave was a sequence of 50 ms rectangular pulses, repeated every 400, 200, and 100 ms (2.5, 5, and 10 Hz, respectively). This approximately comprised the range from 3.3 to 5 Hz implied in the recommendation of 100 ms pulses separated by 100-to-200 ms periods (Lloyd et al., 1999), and included a case of higher frequency. The tenth variant was a visual FCW used as baseline measurement. It was an icon with a red frame and white background, displayed on the lower left corner of the screen. Each type of FCW was tested once by every subject, during one session of approximately 1 hour. The sequence of the FCW variants was different across the subjects, in order to prevent an adaptation bias in the results. 2.3. Braking response The PC of the driving simulator provided a continuous output of vehicle speed, v(t), headway to the leading vehicle, h(t), and angles of the accelerator and brake pedals, α(t) and β(t) respectively. To measure the reaction to the FCW the instant of its activation (t0 ) was also needed. That information was implicit in the PC of the haptic pedal, which provided an output of the angle of the pedal α p (t) during 30 seconds after that instant. The maximum value of the cross-correlation between both pedal angle records (α?α p ) defined the instant t0 when both pedal signals matched, as represented in the scheme of Figure 3. The values of the simulator outputs after t0 were used to define the following parameters: 1. Brake reaction time (BRT): the time that passed from the activation of the FCW to the instant at which the driver pushed the brake pedal: β(t0 + BRT) > 0. 2. Matching speed time (MST): the time from the activation of the FCW to the instant at which the own vehicle reached the speed of the leading vehicle: v(t0 + MST) = 80 km/h. 3. Matching speed deviation (MSD): the variability (standard deviation) of speed in the 5 seconds that followed the instant of matching speed. 4. Headway reduction (HWR): the difference between the headway at the instant of FCW activation and the minimum headway reached while braking: h(t0 ) − hmin . These parameters characterize the braking response of participants: the lower they are, the faster and better the braking action is performed. ANOVA was used to assess if these parameters changed depending on the different conditions of the test. 6

6,08/$725 v, h, β

3&6LP

t0

 

α(t)



'DWD $FTXLVLWLRQ &DUG



α



  WV





αp(t)

 

3&3HGDO

3('$/



  WV

Figure 3: Data synchronization scheme.

2.4. User’s perception In order to assess the sensations caused by the different vibrations, we performed a semantic evaluation, which is a standard technique of Kansei Engineering. From technical documents, magazines and user forums, we gathered a set of 18 adjectives that were repeatedly used to describe vehicle pedals, ADAS, traffic-related warnings and general emotions in automotive contexts (Table 1). Those adjectives comprised most subjective notions that had been evaluated in previous studies, as commented on in the introduction, with the exception of “timeliness”. That concept, however, was not significant in the present study, since the experimental design did not consider changes in the promptness of the FCW. After braking, participants were requested to evaluate the sensations that they had just expe-

Table 1: Adjectives used in the subjective evaluations (original Spanish in brackets).

Pedals soft (blando) hard (duro) stiff (r´ıgido) firm (firme) relaxed (flojo) tensed (tenso) robust (robusto) powerful (potente)

General automotive comfortable (c´omodo) effective (eficaz) pleasant (agradable) innovative (innovador)

7

Warning signal stressing (estresante) reliable (confiable) safe (seguro) stable (estable) alarming (alarmante) annoying (molesto)

rienced, rating the FCW according to the set of adjectives in a 5-point Likert scale, from “totally disagree” to “totally agree”. The evaluation was done aloud, and the supervisor used a different order of the adjectives for each participant. A principal components analysis (PCA) was performed to define the main semantic axes. The scores of these axes were also analysed by an ANOVA in order to define the effect of frequency and amplitude of vibration on the subjective evaluation, just like was done with the objective measurements. The effect of modality (visual vs. vibrotactile) was not included in this analysis, since only the vibrations were subjectively evaluated. 3. Results 3.1. Braking response The primary objective of this study was to assess the effect of the FCW characteristics on the participants’ braking response. However, it has been reported that drivers’ reactions are strongly influenced by the constraints of the car following situation, even more than by the properties of warnings (Tijerina et al., 2000). Therefore, it was necessary to analyze the effect of the relative positions and speeds of the vehicles when the FCW was triggered, in order to assess its potential influence over the results. The risk of collision between two vehicles can be suitably parametrized by TTC. Its estimated value at the moment of the FCW activation was based on the distance between the vehicles when the researcher triggered it, and an assumed speed of 120 km/h. However, since there was a variability in these values, the resulting TTC also varied across the experiments. The average value of TTC was equal to 4.3 s and it was asymmetrically distributed with a standard deviation of 1.9 s, the central 90% cases being between 2.5 s and 8.4 s. TTC had a mild, but significant positive correlation with BRT (Pearson’s ρ = 0.18, p < 0.01), and with HWR (ρ = 0.17, p < 0.01). This confirms that the more critical the situation was, the more intensely participants reacted, in agreement with previous studies (Tijerina et al., 2000). An ANOVA over TTC across the different FCW configurations did not show significant deviations (F(9, 225) = 1.02, p = 0.423). Thus, although the braking response was indeed influenced by the variations of TTC, that variation was evenly distributed in the experimental design, so it was not expected to influence the rest of the analysis. There were 5 of the 300 cases (1.7%) in which the driver failed to match the speed of the vehicle in front and crashed. These cases were excluded from the analysis. Table 2 summarizes the braking response parameters observed for the vibrotactile and visual variants of the FCW (after removing the cases of crash). It can be observed that all the parameters have lower means for the vibrotactile stimuli, the difference being significant for all cases (t(266) between −3.58 and −6.46, p < 0.05). Therefore, braking performance was clearly better with the vibrotactile type of FCW. When the results were analysed as a function of the vibration characteristics, it was found that the lowest averages of all parameters were always achieved at the greatest amplitude (1.60 Nm), and at either the intermediate or the highest frequency (10 Hz for BRT, or 5 Hz for MST, MSD, and HWR). For the other amplitudes and frequencies, the relative positions of the averages varied across the parameters. However, not all the differences were equally significant. BRT was affected by both amplitude (F(2, 58) = 4.20, p < 0.05) and frequency (F(2, 58) = 6.13, p < 0.05), but there was no significant interaction between these two effects (F(4, 116) = 1.90, p = 0.12). A post-hoc 8

Table 2: 10th , 50th , and 90th percentiles of the braking response parameters, for vibrotactile and visual FCW.

FCW 10% 50% 90%

BRT (s) Vibr. Visual 0.30 0.50 0.70 1.00 1.30 1.96

MST (s) Vibr. Visual 2.00 2.28 2.70 3.15 5.92 6.16

1.4

HWR (m) Vibr. Visual 12.71 12.90 20.85 27.55 37.52 47.35

4.5 0.50 Nm 1.05 Nm 1.60 Nm

1.3 1.2

High−BRT

0.50 Nm 1.05 Nm 1.60 Nm 4

1.1

EMM. MST (s)

EMM. BRT (s)

MSD (km/h) Vibr. Visual 1.60 2.30 3.51 3.96 6.23 7.12

1 0.9 0.8

3

Low−BRT

0.7

3.5

0.6 0.5

2.5

5 Freq. (Hz)

2.5

10

5

10

0.50 Nm 1.05 Nm 1.60 Nm 30 EMM. HWR (m)

EMM. MSD (km/h)

5 Freq. (Hz)

35 0.50 Nm 1.05 Nm 1.60 Nm

4.5

4

3.5

3

2.5

25

20

2.5

5 Freq. (Hz)

15

10

2.5

5 Freq. (Hz)

10

Figure 4: Estimated marginal means and their standard errors, resulting from the ANOVA to BRT (top-left), MST (topright), MSD (bottom-left) and HWR (bottom-right).

analysis revealed that BRT was higher for vibrations at 0.50 Nm than for 1.60 Nm, and that 2.5 Hz vibrations also yielded higher BRT values than vibrations at 5 or 10 Hz. Since interactions were not significant, this means that we could define a “High-BRT” group formed by the combinations of amplitudes and frequencies that yielded higher BRT values (0.50 Nm, 2.5 Hz), and a “Low-BRT” group formed by the combinations that yielded lower BRT values (1.60 Nm, 5-10 Hz). The “Low-BRT” group could be regarded as a set of vibrations that achieved faster braking reactions. Figure 4 shows that the BRT means for 1.05 Nm, which were not distinguished from other amplitudes as a group, were aligned with 0.50 Nm at 2.5 Hz (High-BRT), and with 1.60 Nm at 5 or 10 Hz (Low-BRT). On the other hand, MST and MSD were not significantly affected by either amplitude (F(2, 58) = 0.61, p = 0.55 for MST; F(2, 58) = 2.12, p = 0.13 for MSD) or frequency (F(2, 58) = 2.09, p = 0.13 for MST; F(2, 58) = 1.04, p = 0.36 for MSD), nor was there any significant interaction between effects (F(4, 116) = 0.63, p = 0.64 for MST; F(4, 116) = 1.14, p = 0.34 for MSD). 9

Table 3: Estimated marginal means of BRT (s).

0.50 Nm 1.05 Nm 1.60 Nm All 2.5 Hz 1.06† 1.12† 0.69 0.96 5 Hz 0.95 0.58* 0.57* 0.70 10 Hz 0.69 0.63* 0.60* 0.64 All 0.90 0.78 0.62 0.77 † Combinations of the High-BRT group * Combinations of the Low-BRTgroup Table 4: Estimated marginal means of HWR (m).

2.5 Hz 5 Hz 10 Hz All

0.50 Nm 26.8 23.9 22.0 24.3

1.05 Nm 29.2 18.6 26.1 24.7

1.60 Nm 24.7 22.0 20.3 22.4

All 26.9 21.5 22.8 23.9

HWR was affected by frequency (F(2, 58) = 5.57, p < 0.05), but not by amplitude (F(2, 58) = 0.57, p = 0.46). There was also an interaction effect between amplitude and frequency, although it was only marginally significant (F(4, 116) = 2.28, p = 0.06). The post-hoc analysis revealed that, like happened with BRT, the vibrations at the lowest frequency (2.5 Hz) yielded worse results (a higher HWR) than vibrations at higher frequencies. This may be seen in Figure 4 for amplitudes of 0.50 Nm and 1.60 Nm, although this trend was not maintained for 1.05 Nm (this was presumably the cause of the marginal interaction effect). The estimated marginal means of BRT and HWR, the parameters that were significantly affected by the vibration characteristics, are detailed for the different combinations of amplitude and frequency in Tables 3 and 4. These tables show that the average difference in BRT between the High-BRT and Low-BRT groups was 0.5 s, and that HWR at 2.5 Hz was 5.4 m greater than at 5 Hz, and 4.1 m greater than at 10 Hz. 3.2. User’s perception There were three principal components (PC1, PC2, PC3) with an eigenvalue greater than 1, that jointly explained 62.5% of the variance in the participants’ responses. Figure 5 shows the weight of each original adjective upon those components, normalized by a Varimax rotation. The main relations between adjectives and components (with a weight greater than 0.5) were: • PC1 was chiefly related to “hard”, “stiff”, “firm”, “tensed”, “robust”, and (inversely) to “soft” and “relaxed”. Therefore, it may be considered as a component that described a mechanical feature, related to the force perceived by the user. This component explained 22.4% of the variance. • PC2 was chiefly related to “effective”, “innovative”, “reliable”, “safe”, “stable”, and “alarming”. It described a functional feature, related to the efficacy and safety of the FCW, and explained 21.5% of the variance.

10

annoying

alarming

stable

safe

reliable

stressing

innovative

pleasant

PC3

effective

PC2

comfortable

powerful

robust

tensed

firm

stiff

hard

soft

relaxed

PC1 1 0.8 0.6 0.4 0.2 0 -0.2 -0.4 -0.6 -0.8 -1

Figure 5: Weight of the adjectives upon the three principal components.

• PC3 was directly related to “stressing”, “alarming”, “annoying”, and inversely to “comfortable” and “pleasant”. It described a negative ergonomic feature, and explained 18.6% of the variance. The scores of PC1 (the mechanical component) were affected by the vibration amplitude (F(2, 58) = 11.19, p < 0.05), but not by frequency (F(2, 58) = 0.384, p = 0.68), although the interaction between amplitude and frequency did have a significant effect (F(4, 116) = 3.26, p < 0.05). The post-hoc analysis indicated that 0.50 Nm vibrations yielded lower PC1 scores (a weaker perception of force) than greater amplitudes. As may be observed in Figure 6, at 0.50 Nm the lowest frequency (2.5 Hz) also yielded the lowest PC1 scores, although for greater magnitudes this trend was inverted. The scores of PC2 (the functional component) were affected by both amplitude (F(2, 58) = 26.25, p < 0.05) and frequency (F(2, 58) = 9.09, p < 0.05), but not by their interaction (F(4, 116) = 1.10, p = 0.36). According to the post-hoc analysis, the scores of PC2 increased (users perceived a stronger efficacy of the FCW) with both amplitude and frequency, but while there were significant differences across all the tested amplitudes, frequency only yielded differences between 2.5 Hz and the higher values. PC3 (the ergonomic component) behaved just like PC2: there were significant effects of amplitude (F(2, 58) = 35.26, p < 0.05) and frequency (F(2, 58) = 5.09, p < 0.05), but not of their interaction (F(4, 116) = 1.70, p = 0.16). The scores of PC3 (the sensation of displeasure) increased with both amplitude and frequency, with significant differences across all magnitudes, but only between 2.5 Hz and the higher frequencies. 4. Discussion and conclusions 4.1. Interpretation of the results This study validated the efficacy of vibrotactile stimuli for a FCW. Comparing the median values of their response, the reaction of drivers was 0.30 s faster than for a visual warning, they 11

0.5 Nm 1.05 Nm 1.60 Nm

0.6 0.4

0.8

0.4 0.2

0

EMM. PC2

EMM. PC1

0.2

−0.2 −0.4

0 −0.2 −0.4

−0.6

−0.6

−0.8 −1

0.5 Nm 1.05 Nm 1.60 Nm

0.6

−0.8 2.5 Hz

5 Hz Freq. (Hz)

−1

10 Hz

0.8 0.6

2.5 Hz

5 Hz Freq. (Hz)

10 Hz

0.5 Nm 1.05 Nm 1.6 Nm

EMM. PC3

0.4 0.2 0 −0.2 −0.4 −0.6 −0.8

2.5 Hz

5 Hz Freq. (Hz)

10 Hz

Figure 6: Estimated marginal means resulting from the ANOVA to PC1, PC2, and PC3 (from top to bottom and from left to right).

12

matched the speed of the leading vehicle 0.45 s earlier, the deviation of speed afterwards was 0.45 km/h lower and the approximation between the vehicles was 6.7 m shorter. In fact, the ratio of failed cases resulting in crash was also lower with the vibrotactile FCW (1.1% vs. 7.1%). Moreover, we found that variations in the amplitude and frequency of the vibrotactile signal also affected the braking response. In broad terms, this effect coincided with the expected results: the “stronger” and “faster” was the signal, the more effective was the FCW in terms of braking response (the analysed parameters had lower values), so that the weaker and slower vibrations yielded average results similar to the visual FCW. But this general trend was not homogeneous over the whole range of amplitudes and frequencies. The maximum amplitude within the tested range (1.60 Nm) yielded the lowest averages of all the parameters, in agreement with the described trend, but the differences were only significant for BRT, in relation to the result at the lowest amplitude (0.50 Nm). On the other hand, the frequency that yielded the lowest averages alternated between the intermediate and the highest ones (5 Hz and 10 Hz, respectively). The difference between the values of this group and the lowest frequency (2.5 Hz) was significant for BRT and HWR. In both cases 2.5 Hz yielded worse results, in agreement with the described trend. User’s perception was also studied through a semantic analysis. The adjectives that had been found in literature for describing pedals were aligned in the first principal component of the participants’ answers, implying that one of the main aspects perceived by the users was the mechanical quality of the vibration. The comparison of the ratings of this component across the amplitudes and frequencies, revealed that the perception of 0.50 Nm vibrations was different than the perception of greater amplitudes. The former were generally felt softer, weaker and less hard, stiff, etc.; and at that amplitude, such sensation was less intense for 2.5 Hz than for 5 Hz or 10 Hz. Conversely, 1.05 Nm or 1.60 Nm vibrations had not only greater ratings of that component on average, but also an inverted dependence between frequency and that rating (2.5 Hz vibrations were felt harder, stiffer, etc. than “faster” vibrations). The adjectives generally related to automotive industry and warnings were mixed in the other two principal components, which described the functional and the (negative) ergonomic qualities of the vibration. The comparison of the ratings of those components across the vibration amplitudes and frequencies, resulted in a more straightforward and coherent dependence. In agreement with the objective results, the subjective perception of the FCW functionality improved as the amplitude and frequency increased, with significant differences across all amplitudes, but only between 2.5 Hz and higher frequencies. The negative ergonomic perception followed exactly the same pattern, although it was an independent component. These results may be applied to the design of vibrotactile FCW. In order to achieve the best results in performance, the abovementioned facts imply that it may be convenient to set the amplitude of the vibration to the maximum within the tested range (1.60 Nm), and the frequency to the intermediate value (5 Hz). Nevertheless, other combinations may yield a similar performance. The statistical analysis revealed that the so-called “Low-BRT” group of vibrations (1.05-1.60 Nm, 5-10 Hz), to which the proposed combination belongs, has a reaction time significantly shorter (0.5 s of difference) than the “High-BRT” group (0.50-1.05 Nm, 2.5 Hz), but the difference of 0.06 s within the “Low-BRT” group has not been found significant. Likewise, the 1.3 m difference in HWR between 5 Hz and 10 Hz vibrations, has not been found significant in contrast to the 5.4 m difference between 5 Hz and 2.5 Hz. Regarding the subjective experience, increasing either amplitude or frequency would mean increasing both the perception of functionality and discomfort. 13

4.2. Comparison with previous studies and limitations The observed effect of vibration characteristics contrasts with the results of another study of a FCW in the form of a vibrotactile steering wheel, for which the type of vibration was not found to affect the drivers’ response (Tijerina et al., 2000). However, the same study tested the response when the signal was a mono-pulse exerted on the pedal, and there the jerk rate and the duration of the signal did have a significant effect on the braking response. That discrepancy has been considered to result from the fact that a signal displayed on the pedal is, unlike in the steering wheel, intrinsically associated with a braking response. The results of the present study agree with that conclusion. The measured reaction times can be also compared with the results of previous studies. Depending on the type of FCW, the average BRT varied between 0.57 s (for vibration at 1.60 Nm, 5 Hz) and 1.12 s (for 1.05 Nm, 2.5 Hz, similar to the median 1.00 s of the visual FCW). In other studies the total time between the warning and the braking action has been reported in ranges coherent with these results: from 0.62 s to 0.75 s for haptic warnings in a real vehicle (Tijerina et al., 2000), from 0.93 s to 1.12 s for auditory warnings in a simulator (Abe and Richardson, 2006), and from 0.50 s to 0.65 s for visual feedback (brake lights of the leading vehicle in a simulator) (Warshawsky-Livne and Shinar, 2002). Thus, the values gathered in these experiments are coherent with the revised literature. The differences between studies may be due to the different conditions of the experiments. This study confirmed that the constraints of the car following situation influences drivers’ reactions (Tijerina et al., 2000), as BRT and HRW were correlated with the TTC when the warning was triggered. The effect of FCW characteristics could be analyzed, because the variation of TTC was evenly distributed across the experiments, and it did not introduce a bias in the results. But if the speed and headway had been different, or the driving scene had been less constrained, we could have obtained other values within the ranges observed in the literature. The cited studies made a more detailed distinction between the time elapsed until the accelerator pedal was released, and until the brake pedal was depressed. However, the task in this experiment was not to stop the vehicle completely, but to reduce the speed, and the simulator had a set of two pedals that could be simultaneously pressed, so it was possible that the drivers started to brake before releasing the accelerator. Therefore, in this particular study the accelerator release time was not expected to provide relevant information and was omitted from the analysis. The clear advantage of vibrations over visual displays in these experiments supports the benefit of haptic stimuli reported in previous studies which compare haptic and acoustic signals, as cited in the introduction. But these findings do not exhaust the solutions to improve the efficacy of FCW. Particularly, mixed modalities have a great potential for improving the results. Moreover, there are other configurations of amplitude and frequency for vibrotactile stimuli that remain to be studied. This study provided general trends of the effect of those variables, which can be used to estimate how it may vary within the tested ranges. But it must be considered that such trends may change for values far from the experimented range. The minimum amplitude, 0.50 Nm, was 7.2% of the greatest torque exerted by the pedal; therefore, it was near the discrimination threshold of forces on the foot, which has been suggested to be a Weber fraction of 7% (Southall, 1985). The ratios of braking failures indicate that even those weak vibrations were successfully noticed by most participants, but lower amplitudes would likely result in very poor performance. On the other hand, the maximum amplitude (1.60 Nm, 23.2% of the top pedal torque) yielded the best performance averages, but if the amplitude of the vibration reaches excessive values in 14

relation to the force exerted to depress the pedal, it can be expected to interfere with the control of acceleration. Important departures in frequency might also affect the observed trends in different ways, as well as the subjective evaluation. The applicability of the results may be also conditioned to other characteristics of the FCW signal, that were fixed in this study: the shape of the vibration wave, which was a sequence of rectangular pulses, the duration of the signal, which in this case was maintained until the drivers stabilized their speed, or the anticipation of the warning, which varied around an average value of TTC = 4.3 s, but was not strictly controlled. In fact, alarm promptness has been found to have an important influence trust in the system: as an overall trend, as alarms are provided earlier driver trust is higher, although when driving demand is low, trust is also high even for relatively late alarms, and vice versa (Abe and Richardson, 2006). One of the important contributions of this study is the subjective characterization of vibrotactile stimuli, through Kansei methodologies. Although previous studies have also gathered subjective information from questionnaires, this technique allowed to define the main general concepts that characterize the perception of pedal vibrations, covering the semantic fields used in the revised studies (with the exception of the notion of “timeliness” (Abe and Richardson, 2006), which was not an element to be analysed), but going beyond the particular terminology used in the questionnaire. It should be mentioned that, since the study was conducted with Spanish test persons, the terms that were used in the semantic analysis were not the English adjectives commented on through the article, but their Spanish equivalents, as reflected in Table 1. Nevertheless, the risk of cultural bias is limited, because the principal component analysis yielded three general concepts that were not associated to particular words, and can be easily transferred to semantic fields of other languages. 4.3. Potential application In conclusion, it has been found that vibrotactile stimuli have an important potential to improve the efficacy of FCW and there are certain types of vibration that achieve better results. Moreover, the perception of such efficacy is generally aligned with the objective results, but also with the sensation of discomfort. Therefore, the configuration of vibrotactile warnings must be chosen with caution, in order to balance efficacy and satisfaction. In relation to this, assessments of user acceptance would contribute to define if the improvement of safety makes up for the possible nuisance, like has been done with active accelerator pedals for speed management in field studies (Adell and V´arhelyi, 2008; Adell et al., 2008), or for a congestion assistant in a simulator (Brookhuis et al., 2008; van Driel et al., 2007). There are, in addition, other behavioural aspects that can be considered to improve the efficacy of FCW. Adaptive, more effective warnings could be configured with information of the driver’s personal behaviour, like sensation seeking, preferred headway or individual brake reaction time (Jamson et al., 2008), or even situation-specific behaviour, like aggresivity caused by anger, which is associated to greater risk in driving (Stephens and Groeger, 2009). A risk of “complacency” has also been reported for users accustomed to highly reliable FCW (Maltz and Shinar, 2007), so perfect single-stage warnings alone are not the best solution to enhance safety, while graded warnings may provide greater safety margins without habituation effects (Lee et al., 2004), and continuous graded feedback may enhance the situation awareness, thus reducing the need of warnings (Abbink et al., 2008). Future work should be done, introducing some factors that were excluded from this study (combining haptic and other modalities, different driving scenarios, alarm anticipation, etc.), and 15

taking into account the abovementioned behavioural aspects, as well as user acceptance, which is the key for a successful safety product. Acknowledgements This study has been done within the framework of the project MARTA, led by FICOSA INTERNATIONAL S.A. and funded by the Centro para el Desarrollo Tecnol´ogico Industrial (CDTI) for the 3rd CENIT Programme, as a part of the INGENIO 2010 Programme of the Spanish Government. References 2007/78/EC, 6 February 2007. Commission recommendation of 22 december 2006 on safe and efficient in-vehicle information and communication systems: update of the european statement of principles on human machine interface. Tech. rep., Commission of the European Communities. Abbink, D. A., Boer, E. R., Mulder, M., June 4-6 2008. Motivation for continuous haptic gas pedal feedback to support car following. In: 2008 IEEE Intelligent Vehicles Symposium. Eindhoven, The Netherlands, pp. 283–290. Abe, G., Richardson, J., 2006. Alarm timing, trust and driver expectation for forward collision warning systems. Applied Ergonomics 37 (5), 577–586. Ackermann, J., B¨unte, T., Odenthal, D., 1999. Advantages of active steering for vehicle dynamics control. In: 32nd International Symposium on Automotive Technology and Automation. pp. 263–270. Adell, E., V´arhelyi, A., 2008. Driver comprehension and acceptance of the active accelerator pedal after long-term use. Transportation Research Part F: Traffic Psychology and Behaviour 11 (1), 37–51. Adell, E., V´arhelyi, A., Hj¨almdahl, M., 2008. Auditory and haptic systems for in-car speed management - a comparative real life study. Transportation Research Part F: Traffic Psychology and Behaviour 11 (6), 445–458. Amditis, A., Polychronopoulos, A., Bekiaris, E., 2007. Real-time traffic and environment risk estimation for the optimisation of human-machine interaction. In: Cacciabue, P. C. (Ed.), Modelling driver behaviour in automotive environments: critical issues in driver interactions with intelligent transport systems. Springer, pp. 379–399. Bar-Gera, H., Shinar, D., 2005. The tendency of drivers to pass other vehicles. Transportation Research Part F: Traffic Psychology and Behaviour 8 (6), 429–439. Borro, D., Savall, J., Amundarain, A., Gil, J. J., Garc´ıa-Alonso, A., Matey, L., November-December 2004. A large haptic device for aircraft engine maintainability. IEEE Computer Graphics and Applications 24 (6), 70–74. Brookhuis, K. A., de Waard, D., 2004. Adas’ acceptance and effects on behaviour: the consequences of automation. In: Underwood, G. (Ed.), Traffic and Transport Psychology: Theory and Application : Proceedings of the ICTTP 2004. Elsevier, pp. 273–278. Brookhuis, K. A., van Driel Cornelie, J. G., Hof, T., van Arem, B., Hoedemaeker, M., 2008. Driving with a congestion assistant; mental workload and acceptance. Applied Ergonomics In Press, Corrected Proof. Frisoli, A., Avizzano, C. A., Bergamasco, M., May 21-26 2001. Simulation of a manual gearshift with a 2 dof forcefeedback joystick. In: 2001 IEEE International Conference on Robotics and Automation. Seoul, Korea, pp. 1364– 1369. Hirst, S., Graham, R., 1997. The format and presentation of collision warnings. L. Erlbaum Associates Inc., Hillsdale, NJ, USA, pp. 203–219. Hj¨almdahl, M., V´arhelyi, A., 2004. Speed regulation by in-car active accelerator pedal: Effects on driver behaviour. Transportation Research Part F: Traffic Psychology and Behaviour 7 (2), 77–94. Ho, C., Tan, H. Z., Spence, C., 2005. Using spatial vibrotactile cues to direct visual attention in driving scenes. Transportation Research Part F: Traffic Psychology and Behaviour 8 (6), 397–412. Hoffman, J. D., Lee, J. D., Hayes, E. M., 2003. Driver preference of collision warning strategy and modality. In: 2nd International Driving Symposium on Human Factors in Driver Assessment, Training and Vehicle Design. Horiguchi, A., Suetomi, T., 1995. A kansei engineering approach to a driver/vehicle system. International Journal of Industrial Ergonomics 15 (1), 25–37. Jamson, A. H., Lai, F. C. H., Carsten, O. M. J., 2008. Potential benefits of an adaptive forward collision warning system. Transportation Research Part C: Emerging Technologies 16 (4), 471–484. Jordan, N., Franck, M., Jean-Michel, H., 2007. Lateral control support for car drivers: a human-machine cooperation approach. In: Proceedings of the 14th European Conference on Cognitive ergonomics. ACM, New York, NY, USA, pp. 249–252.

16

Lee, J. D., Hoffman, J. D., Hayes, E., 2004. Collision warning design to mitigate driver distraction. In: CHI ’04: Proceedings of the SIGCHI conference on Human factors in computing systems. ACM, New York, NY, USA, pp. 65–72. Li, M., Liu, Y.-H., May 15-19 2006. Haptic modeling and experimental validation for interactive endodontic simulation. In: 2006 IEEE International Conference on Robotics and Automation. Orlando, Florida, USA, pp. 3292–3297. Lloyd, M. M., Wilson, G. D., Nowak, C. J., Jr., A. C. B., 1999. Brake pulsing as haptic warning for an intersection collision avoidance countermeasure. Transportation Research Record: Journal of the Transportation Research Board 1694, 34–41. Madhani, A. J., Niemeyer, G., Salisbury, J. K., October 1998. The black falcon: A teleoperated surgical instrument for minimally invasive surgery. In: 1998 IEEE/RSJ International Conference on Intelligent Robots and Systems. Victoria B.C., Canada, pp. 936–944. Maltz, M., Shinar, D., 2007. Imperfect in-vehicle collision avoidance warning systems can aid distracted drivers. Transportation Research Part F: Traffic Psychology and Behaviour 10 (4), 345–357. Martens, M., van Winsum, W., 2001. Effects of speech versus tactile driver support messages on driving behavior and workload. SAE Technical Papers 2001-06-0080, SAE International. Mulder, M., Mulder, M., van Paassen, M. M., Abbink, D. A., 2008. Haptic gas pedal feedback. Ergonomics 51 (11), 1710–1720. Nagamachi, M., 1995. Kansei engineering: A new ergonomic consumer-oriented technology for product development. International Journal of Industrial Ergonomics 15 (1), 3–11. Southall, D., 1985. The discrimination of clutch-pedal resistances. Ergonomics 28 (9), 1311–1317. Stanley, L. M., 2006. Haptic and auditory cues for lane departure warnings. Human Factors and Ergonomics Society Annual Meeting Proceedings, Surface Transportation 4, 2405–2408. Steele, M., Gillespie, R. B., 2001. Shared control between human and machine: Using a haptic steering wheel to aid in land vehicle guidance. In: Proceedings of the Human Factors and Ergonomics Society 45th annual meeting. pp. 1671–1675. Stephens, A. N., Groeger, J. A., 2009. Situational specificity of trait influences on drivers’ evaluations and driving behaviour. Transportation Research Part F: Traffic Psychology and Behaviour 12 (1), 29–39. Suzuki, K., Jansson, H., 2003. An analysis of driver’s steering behaviour during auditory or haptic warnings for the designing of lane departure warning system. JSAE Review 24 (1), 65–70. Tideman, M., van der Voort, M. C., van Houten, F. J. A. M., June 14-17 2004. Design and evaluation of a virtual gearshift application. In: 2004 IEEE Intelligent Vehicles Symposium. Parma, Italy, pp. 465–470. Tijerina, L., Johnston, S., Parmer, E., Pham, H. A., Winterbottom, M. D., Barickman, F. S., 2000. Preliminary studies in haptic displays for rear-end collision avoidance system and adaptive cruise control system applications. Tech. rep., U.S. Department of Transportation. National Highway Traffic Safety Administration, Washington D.C. van Driel, C. J. G., Hoedemaeker, M., van Arem, B., 2007. Impacts of a congestion assistant on driving behaviour and acceptance using a driving simulator. Transportation Research Part F: Traffic Psychology and Behaviour 10 (2), 139–152. van Erp, J. B. F., van Veen, H. A. H. C., 2004. Vibrotactile in-vehicle navigation system. Transportation Research Part F: Traffic Psychology and Behaviour 7 (4-5), 247–256. Warshawsky-Livne, L., Shinar, D., 2002. Effects of uncertainty, transmission type, driver age and gender on brake reaction and movement time. Journal of Safety Research 33 (1), 117–128.

17