Neuromodulated Control of An Autonomous Robot Prince Akimul Dr. Biswanath Samanta Department of Mechanical Engineering
Robot Control Implementation A neural network model was proposed to emulate the behavioral characteristics of biological organism using neuromodulation on an autonomous robot platform [2]. The neuronal model along with the Matlab code was adapted for the present work to test the neuromodulated behavior of the autonomous iRobot Create. The robot was equipped with relatively inexpensive ping ultrasound sensors in place of the expensive laser rangefinder (URG-04-LX) used in [2].The robot was run using Matlab in Windows on a notebook PC placed on top of the robot as shown in Fig. 1. Three of the built-in sensors, namely, bump, dock beam, and battery state of iRobot Create were used in this study. In addition to these, three ultrasonic ping sensors were mounted in the front, left and right side of the robot for measuring the robot distance from a nearby object like wall or any obstacle. The entire neuronal model’s structure was used as a control algorithm for the robot and the control algorithm was written using the Matlab toolbox for iRobot. The ultrasonic rangefinder sensors were initiated by the Sketch program compiled and run by an Arduino Uno board. The ping sensor data were sent to Matlab workspace from Arduino board through serial interfacing between Matlab and Arduino.
Fig. 1: The iRobot Create with Ping sensors and laptop
Stay Home
5
Behavioral States
6
Find Home
4
Explore Obj
3
Open Field
2
Wall Follow
1 0
50
100 150 200 Experiment Run Time(s)
250
300
Risk Aversive mode: State transition of the Robot Behavioral State Transiton
Behavioral States
Methodology
Behavioral State Transiton Leave Home
Leave Home
6
Stay Home
5
Find Home
4
Explore Obj
3
Open Field
2
Wall Follow
1 0
100 200 Experiment Run Time(s)
300
Risk Taking mode: State transition of the Robot Fig. 2: The structure of neural network model. The connections between the layer of event neurons and state neurons and the intrinsic connections amongst the state neurons are not shown. Experimental Results A series of experiments were run in a lab studio environment where many tables, chairs and other solid objects were kept. Initially, 5 minutes of experiments were carried out to see the robot’s behavioral performance under three different running conditions. 1) Risk aversive behavior: Bumps were treated as potentially harmful by connecting Bump event neurons to the 5-HT neuron. 2) Risk taking behavior: Bumps were treated as novel and interesting by connecting Bump event neurons to the DA neuron. 3) Distracted behavior: The second condition was repeated with the ACh/NE neurons kept always active (activity value =1).
Behavioral State Transiton
Behavioral States
The poster presents a control approach based on vertebrate neuromodulation and its implementation on an autonomous robot platform. A neural network is used to model the neuromodulatory function for generating context based behavioral responses to sensory signals. The implementation of the neuronal model on a relatively simple autonomous robot illustrates its interesting behavior adapting to changes in the environment.
The 3-layer network structure of Fig. 2 has been used to study how efficiently it would perform to control an autonomous robot’s behavior. The neurons in three layers represent events from sensory signals, neuromodulatory system and behavior states respectively. The first layer of neurons, termed as events neurons, indicates the incidents happening in the real world environment for the robot. The entire experiment was run to test how the robot would respond if any of events on the first layer takes place. This network structure is designed in such a way that it is capable of accommodating several events taking place in the real world environment at the same time [2].
Behavioral states
Introduction
Leave Home 6 Stay Home
5
Fi nd Home
4
Expl ore Obj 3 Open Fi el d
2
Wal l Fol l ow 1
0
20
40 60 80 Experiment Run Time(s)
100
120
Distracted mode: State transition of the Robot [1] J. L. Krichmar, “A neurorobotic platform to test the influence of neuromodulatory signaling on anxious and curious behavior,” Frontiers in Neurorobotics, vol. 6, 5, pp. 1-16, 2013. [2] J. L. Krichmar, “A biologically inspired action selection algorithm based on principles of neuromodulation,” IJCNN, 2012 , pp. 1-8. [3]J. L. Krichmar, "The neuromodulatory system – a framework for survival and adaptive behavior in a challenging world," Adaptive Behavior, vol. 16, pp. 385-399, 2008.
