Single-Leg Balancing of Humanoid Robot Using Five-Particle Model Predictive Control

Maintaining balance during single-leg standing is a fundamental yet inherently challenging task for humanoid robots, with the difficulty further amplified under external disturbances. Inspired by the coordination of upper and lower limbs in human balance control, this paper presents a unified control framework that integrates a five-particle model with Model Predictive Control and Whole-Body Control, termed 5P-MPC + WBC. The five-particle abstraction models the torso, arms, and legs as individual dynamic components, enabling more accurate CoM estimation and whole-body trajectory planning. The proposed controller leverages coordinated arm and leg motions to enhance disturbance rejection. Simulations on the 23-DoF Booster T1 humanoid robot demonstrate that our method outperforms conventional 3P-MPC and only WBC approaches in both stability and recovery speed, especially under high-impact disturbances. The controller operates in real-time and offers a promising solution for robust humanoid balance in dynamic environments.