A Model of Operant Conditioning for Adaptive Obstacle Avoidance
Coronado, Juan Lopez
MetadataShow full item record
We have recently introduced a self-organizing adaptive neural controller that learns to control movements of a wheeled mobile robot toward stationary or moving targets, even when the robot's kinematics arc unknown, or when they change unexpectedly during operation. The model has been shown to outperform other traditional controllers, especially in noisy environments. This article describes a neural network module for obstacle avoidance that complements our previous work. The obstacle avoidance module is based on a model of classical and operant conditioning first proposed by Grossberg ( 1971). This module learns the patterns of ultrasonic sensor activation that predict collisions as the robot navigates in an unknown cluttered environment. Along with our original low-level controller, this work illustrates the potential of applying biologically inspired neural networks to the areas of adaptive robotics and control.