Acquisition of a Destination Path Accompanied by Obstacle Avoiding Actions for Quadruped Robots Using Neural Network

Abstract

In the obstacle avoidance of a legged type robot, it is not necessary to avoid all of obstacles by only turning, because it can get over or stride some of them, depending on the obstacle configuration and the state of the robot, unlike a wheel type or a crawler type robot. It is thought that the mobility efficiency to the destination is improved by getting-over or striding. In this paper, a neural network (NN) is used to determine the action of a quadruped robot in the obstacle avoidance path by using information on the destination, the obstacle configuration, and the robot's self-state. The design parameters of NN are adjusted by genetic algorithm (GA) offline. The effectiveness of the present method is proved through an actual experiment.