Walking Gait Learning for “T-FLoW” Humanoid Robot Using Rule-Based Learning

摘要

This work presents a fast and simple learning algorithm for humanoid robot walking gait cases. The standard method of reinforcement learning takes too much time to learn a stable walking gait. Thus, we propose a rule-based learning method that has never been used in this kind of walking gait learning case. We implement our method in a simplified TFLoW humanoid robot model in simulation software CoppeliaSim. The result shows by using our proposed method, T-FLoW humanoid robot can walk for 200 steps after taking the learning process for about 800 episodes and has a better walking performance than the classical pattern generation for planning a walking gait motion.