Legged animals have the flexibility to change their gait locomotion pattern strategies under different environments and walking speeds, which is important for efficient locomotion. However, legged locomotion is currently studied as a fixed set of pre-programmed gait patterns, which limits locomotor flexibility. Moreover, the process of designing multiple coping strategies for multiple tasks is cumbersome. In this study, we introduce a novel framework for quadrupedal locomotion learning, enabling robots to swiftly and adaptably transition between gait patterns tailored to various environments. In addition, we introduce an environmental feedback framework to improve the quadrupedal robot's ability to perceive external environmental features. We obtain a strategy network through a two-stage training and deploy it to real robots. Finally, a multi-strategy controller is obtained, which enables the quadruped robot to realize flexible strategy shifts and, moreover, efficient traversal of various complex terrain environments.
Multi-strategy Central Pattern Generator and Reinforcement Learning Integration for Quadruped Locomotion
Lect. Notes Electrical Eng.
International Conference on Machine Learning, Cloud Computing and Intelligent Mining ; 2024 ; Shennongjia, China August 08, 2024 - August 11, 2024
Proceedings of the 3rd International Conference on Machine Learning, Cloud Computing and Intelligent Mining (MLCCIM2024) ; Chapter : 15 ; 164-176
2025-03-22
13 pages
Article/Chapter (Book)
Electronic Resource
English
Animal-Like Eye Vision Assisted Locomotion of a Quadruped Based on Reinforcement Learning
Springer Verlag | 2023
|A Novel Multi-Gait Strategy for Stable and Efficient Quadruped Robot Locomotion
ArXiv | 2024
|Development of quadruped walking locomotion gait generator using a hybrid method
IOP Institute of Physics | 2013
|