A Centroidal Dynamics-Based MPC Framework for Agile Bipedal Walking

Model predictive control (MPC) has demonstrated excellent performance in real-time motion control for bipedal robots. However, the substantial computational burden imposed by high-fidelity dynamic models requires simplification in practical applications. With the rapid advancement of computational capabilities, the selection of appropriate simplified models has become a critical research focus. This paper provides a comprehensive review of two prevalent dynamic models in MPC-the single rigid body model and the full-body dynamic model-and proposes a novel centroidal dynamics-based model that integrates the advantages of both. The mathematical principles of these three models in MPC are briefly introduced, and their control performance is analyzed through simulation experiments. The strengths and limitations of each model are summarized, aiming to contribute to the selection of dynamic models for MPC algorithms in bipedal robots.