Orientation-Aware Model Predictive Control with Footstep Adaptation for Dynamic Humanoid Walking

Abstract

This paper proposes a novel model predictive control (MPC) for humanoid locomotion that reasons about orientation dynamics and footstep placement in a unified optimization framework. This work employs the augmented single rigid body model (aSRBM) to enable the MPC to leverage stepping strategy and orientation dynamics simultaneously, hence the name orientation-aware MPC (OA-MPC). Since step location is part of the decision variables, this MPC ensures that foot placement satisfies the kinematic reachability constraint and conforms to terrain slope. The OA-MPC produces the desired body pose, ground reaction wrench, and foot swing targets that are tracked by a task-space controller (TSC), which utilizes the full-order dynamics of the humanoid and authorizes arm movements. The proposed control framework is suitable for real-time execution since both MPC and TSC are transcribed as quadratic programs. Simulation investigations show that the OA-MPC is more robust against external torque disturbance compared to controllers using the point mass model, especially when the torso undergoes large angular excursion. The OA-MPC also enables the MIT Humanoid to traverse the wave field.