Gait and Trajectory Optimization by Self-Learning for Quadrupedal Robots with an Active Back Joint

Abstract

This paper presents an efficient technique for a self-learning dynamic walk for a quadrupedal robot. The cost function for such a task is typically complicated, and the number of parameters to be optimized is high. Therefore, a simple technique for optimization is of importance. We apply a genetic algorithm (GA) which uses real experimental data rather than simulations to evaluate the fitness of a tested gait. The algorithm actively optimizes 12 of the robot's dynamic walking parameters. These include the step length and duration and the bending of an active back. For this end, a simple quadrupedal robot was designed and fabricated in a structure inspired by small animals. The fitness function was then computed based on experimental data collected from a camera located above the scene coupled with data collected from the actuators' sensors. The experimental results demonstrate how walking abilities are improved in the course of learning, while including an active back should be considered to improve walking performances.