Gait Self-learning for Damaged Robots Combining Bionic Inspiration and Deep Reinforcement Learning

• Published in: Chinese Control Conference pp. 3978 - 3983
• Main Authors: Zeng, Ming, Ma, Yu, Wang, Zhijing, Li, Qi
• Format: Conference Proceeding
• Language: English
• Published: Technical Committee on Control Theory, Chinese Association of Automation 26.07.2021
• Subjects: Aerospace electronics; Biological system modeling; bionic inspiration; damaged robot; deep reinforcement learning; Gait self-learning; Legged locomotion; Reinforcement learning; Robot control; Robots; Training
• ISSN: 1934-1768
• DOI: 10.23919/CCC52363.2021.9549968

The gait self-learning for the Hexapod Robot in the damaged state is very important to improve its survivability in complex environments. Aiming at the damage of a Hexapod Robot with broken legs, this paper proposes a gait self-learning method for the damaged robot based on bionic inspiration and deep reinforcement learning. Due to various damage states of robots, it is difficult to accurately model the complex damage state. Using deep reinforcement learning can better solve this kind of model-free robot control problem. Besides, inspired by the moving gaits of the hexapod, the motion range of each leg joint of the Hexapod Robot is constrained, which further reduces the search range of the action space. The experimental results show that compared with the single deep reinforcement learning method under the same training episodes, the gaits generated by the proposed method are more adaptable and efficient.