Neuroevolution and NEAT at Evolving Neurocontrollers for Obstacle Avoidance for a Robot Arm [Articol]

Abstract

This paper presents a comparative analysis of Neuroevolution and NEAT algorithms for evolving neural controllers capable of evading obstacles. The fitness criterion is the number of time steps a robot is able to avoid a moving obstacle that increases in speed over time. The experiments simulate the WLkata Mirobot, a small 6-joint robot, focusing on obstacle evasion using forward kinematics. The results show the average number of generations required by each algorithm to achieve a certain fitness and analyze the average fitness per generation. This comparison offers insights into the effectiveness and efficiency of each algorithm in optimising robotic obstacle avoidance.