Real-time MPPI Control of a 3DoF Leg System With Contact Constraints Projected Into Whole-body Dynamics

This study proposes a whole-body Model Predictive Path Integral (MPPI) approach for unified motion planning and control of a 3-Degree-of-Freedom (DoF) leg system. High-potential dynamics in humanoid robots present significant control challenges, but Whole-Body Dynamics (WBD) enables physically consistent and dynamically rich motions suitable for real-world tasks without restrictive assumptions. Recent approaches combine WBD with Model Predictive Control (MPC) to achieve real-time planning and disturbance responsiveness, capable of generating more stable and agile motions. However, these methods face computational limitations in real-time applications. This study addresses these issues by utilizing MPPI, which leverages sampling-based optimization to generate stable and agile motions. Simulations verify that the proposed method effectively integrates full-body dynamics and supports real-time motion planning and control, with potential applications in dynamic locomotion tasks.