Adaptive behavior of evolutionary neurocontroller for obstacle avoidance mobile robot

Abstract

This research aims to analyze the adaptive behavior of an artificial neural network-based controller (neurocontroller) for an obstacle avoidance wheeled mobile robot. The neurocontroller of interest is the best of an ecosystem of neurocontrollers that evolved according to the genetic algorithm. The chromosome of each individual neurocontroller is simply the binary weights of the connections, and the fitness function is a simple rule for obstacle avoidance behavior. To limit the hardware requirements, the neurocontroller’s evolution is simulated in a computer, and the best neurocontroller behavior is analyzed in the simulation and in an actual mobile robot. Albeit a simple chromosome, the resulting best neurocontroller showed an obstacle avoidance behavior in both simulation and in the actual mobile robot, regardless of the positions of obstacles in the environment. The analysis of the chromosome after evolution also found that not all of the sensors are actually needed for the obstacle avoidance behavior.