An Universal Dynamic Model Predictive Control Framework for Quadruped Robot Locomotion

Abstract

Quadruped Robots require the capability to traverse complex and demanding surroundings found in natural landscapes, urban areas, and industrial facilities. However, achieving efficient and flexible control of these systems remains an ongoing challenge. In this paper, we present a novel Universal Dynamic Model Predictive Control (UDMPC) Framework designed specifically for quadruped robot locomotion, aiming to address the controller design problems caused by the diversity of quadruped robots and the complexity of terrains. Even within intricate terrains, this framework guarantees accurate tracing of reference velocity commands, augmenting locomotion prowess. The MPC controller’s parameters are dynamically fine-tuned through reinforcement learning, enhancing control reliability. Our proposed approach was subjected to rigorous testing and evaluation using the Go1 quadruped robot model within a simulation environment. The findings showcased its exceptional dynamic adaptability, surpassing fixed-parameter controllers. Notably, this work considerably enhances command tracking precision and stability.