A Generalist Agent Learning Architecture for Versatile Quadruped Locomotion

Yanyun Chen1    Ran Song1    Jiapeng Sheng1,2    Xing Fang1    Wenhao Tan1    Yibin Li1    Wei Zhang1
1Shandong University          2Tencent Robotics X Submitted to IEEE Transactions on Automation Science and Engineering

Abstract

Quadrupeds can generate various motor behaviors with the muscle synergies activated by the central nervous system. However, versatile locomotion for quadruped robots remains challenging due to the complexity of the high-dimensional limb dynamics with many physical constraints. Current approaches typically apply a dedicated policy or controller for each motor behavior, which requires the optimization of a large number of parameters and is not bionically sound. Inspired by the CNS and the muscle synergies of quadruped animals, we propose a Generalist Agent Learning Architecture (GALA) to learn diverse motor behaviors simultaneously with a single policy network for quadruped locomotion. GALA significantly decreases the number of trainable parameters while producing appropriate motor behaviors by simply reactivating the generalist policy based on different sensory feedback and commands at run time. We experimentally analyze and demonstrate the versatile locomotion delivered by GALA on both simulated and real quadruped robots in various environments. The code is available at https://github.com/vsislab/GALA.



Video S1: Performance of recovery behavior


Video S2: Performance with randomized commands


Video S3: Demonstration in training sites

Video S4: Demonstration of 7 motor behaviors

Video S5: Demonstration in outdoor environments

Video S6: Demonstration in a variety of indoor and outdoor environments

Video S7: Combinations of different behaviors in both simulated and real world environments