Emergency-response locomotion of hexapod robot with heuristic reinforcement learning using Q-learning

Abstract

The locomotion of legged robot is often controlled by predefined gaits, and this approach works well when all joints and motors are operating normally. However, walking legged robots usually have high risk of being damaged during operation, causing the breakdown of the robotic joints. In this paper, we introduce a reinforcement learning based approach for the legged robot to generate real-time locomotion response to the emergence of locomotion breakdown. Our approach detects the functionality of the available joints, substitutes the pre-defined gaits with proper gait function accordingly, and upgrades the gait-generation function by Q-Learning for the proper locomotion.