Dynamic Gaits Locomotion for Quadrupedal Robots With Gait Phase Decomposition

Abstract

Quadruped robots have witnessed notable progress in motion performance in complex terrains, especially in the fields of terrain adaptation and locomotion robustness. Beyond these aspects, dynamic gaits inspired by the agile movements of quadruped mammals in nature offer a promising area of exploration to enhance both locomotion efficiency and motion robustness in quadruped robots. While existing research may still be in its early stages, considerable technical difficulties remain. For instance, galloping is a highly agile and efficient gait, yet it involves complex foot-ground interactions and locomotion instability. This article proposes a novel gait phase decomposition model predictive control (GPD-MPC) to guarantee fast and efficient dynamic gaits compared with the current existing methods. First, GPD models are derived to segment a stride into distinct stance phases, extracting attitude variation tendencies through decomposing the complex contact sequences for stabilizing dynamic gait motion. Then, an optimal controller is built by integrating these models to generate adaptive ground reaction forces. Finally, by combining GPD-based control with MPC, stable dynamic locomotion and retard motion oscillation can be achieved. Comprehensive simulations and hardware experimental investigations have been conducted based on the Unitree Go2 quadruped robot to perform stable and continuous transverse galloping and bounding on different roads. The simulated and experimental results indicate that the proposed GPD-MPC method can reduce the body’s attitude oscillations of the quadruped robot, enabling fast, and considerably stable dynamic gaits compared with the MPC-only method.