A TD-Learning Based Bionic Cerebellar Model Controller For Humanoid Robots

• Published in: Procedia computer science Vol. 209; pp. 132 - 139
• Main Authors: Li, Honghui, Liu, Rong, Wang, Yongxuan, Liu, Yin, Chen, Yaru, Wang, Jiaxing, Gu, Jason
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 2022
• Subjects: Cerebellum model; Gait control; Humanoid robot; Reinforcement learning; Cerebellum model; Gait control; Reinforcement learning; Humanoid robot
• ISSN: 1877-0509; 1877-0509
• DOI: 10.1016/j.procs.2022.10.107

The cerebellum is a crucial component of the human body that plays a vital role in human walking. To design a robot gait controller by referring the working mechanism of the cerebellum is one of the hotspots in the bionic control field. This paper designs a bionic cerebellar motion control model to control the slope gait of a humanoid robot. The cerebellum model refers to the connection method between neurons in a human cerebellum, and expresses from a bionic perspective how the neurons in the cerebellum process external information and generate control commands during walking. Inspired by how human walking is learned, this model employs reinforcement learning in the learning process of the bionic cerebellar model. A corresponding simulation environment is also designed to train and test the cerebellar control model's effectiveness when regulating a robot's slope walking stability. The simulation experimental results demonstrate that the cerebellum model can achieve stable control of the walking motion of the humanoid robot after training, verifying its effectiveness, and laying a foundation for further realization of human-like artificial intelligence.