Reduced Model Predictive Control Toward Highly Dynamic Quadruped Locomotion

Abstract

Controlling quadruped robots during dynamic motions presents significant challenges due to constraints on ground reaction forces and the inherent complexity of their dynamics. Model predictive control (MPC) has shown promise in addressing these challenges. However, the performance of MPC strongly relies on the accuracy and complexity of the model, making the modeling process critical for dynamic locomotion control. This paper introduces a novel approach using the reduced single rigid body model (SRBM) and an associated MPC for achieving high-frequency control—crucial for highly dynamic locomotion. The reduced SRBM is derived by isolating the key components responsible for robot balance from the full SRBM, reducing model complexity without compromising control performance. Additionally, the planar kinematics is developed that considers the motions neglected in the reduced model. This enables the design of foot trajectories that facilitate omni-directional motion and yaw control. To validate the proposed method, computer simulations are conducted under various scenarios. The simulations demonstrate that the quadruped robot can achieve galloping speeds of up to 7 m/s while remaining stable even when subjected to a lateral disturbance of 200 N.