Obstacle Avoidance for Quadruped Robots Using a Neural Network

Abstract

Legged mobile robots, which differ from wheeled and crawler, need not avoid all obstacles by altering the path in the obstacle avoidance task. Because, legged mobile robots can get over or stride some obstacles, depending on the obstacle configuration and the current state of the robot. Legged mobile robots muse have suitable motion for each leg. We propose body motion control of a quadruped robot using a neural network (NN) for an obstacle avoidance task. Each leg motion is calculated by robot kinematics using body motion from the NN. NN design parameters are tuned off-line by a genetic algorithm (GA). Effectiveness of the present method is proved through an experiment.