Whole-body MPC for highly redundant legged manipulators: experimental evaluation with a 37 DoF dual-arm quadruped

TL;DR

The paper tackles loco-manipulation with highly redundant legged platforms by presenting a real-time whole-body MPC framework that incorporates full kinematics and centroidal dynamics. It demonstrates a two-loop pipeline where the MPC provides joint references to low-level impedance controllers, eliminating the need for an instantaneous whole-body controller. Key contributions include experimental validation on CENTAURO with 37 DoFs, showing heavy object manipulation and dynamic stepping, and achieving replanning frequencies in the tens of Hz. The work significantly advances predictive, whole-body coordination for complex, highly redundant legged manipulators and lays a path toward broader real-world loco-manipulation applications.

Abstract

Recent progress in legged locomotion has rendered quadruped manipulators a promising solution for performing tasks that require both mobility and manipulation (loco-manipulation). In the real world, task specifications and/or environment constraints may require the quadruped manipulator to be equipped with high redundancy as well as whole-body motion coordination capabilities. This work presents an experimental evaluation of a whole-body Model Predictive Control (MPC) framework achieving real-time performance on a dual-arm quadruped platform consisting of 37 actuated joints. To the best of our knowledge this is the legged manipulator with the highest number of joints to be controlled with real-time whole-body MPC so far. The computational efficiency of the MPC while considering the full robot kinematics and the centroidal dynamics model builds upon an open-source DDP-variant solver and a state-of-the-art optimal control problem formulation. Differently from previous works on quadruped manipulators, the MPC is directly interfaced with the low-level joint impedance controllers without the need of designing an instantaneous whole-body controller. The feasibility on the real hardware is showcased using the CENTAURO platform for the challenging task of picking a heavy object from the ground. Dynamic stepping (trotting) is also showcased for first time with this robot. The results highlight the potential of replanning with whole-body information in a predictive control loop.

Figures (7)

• Figure 1: The dual-arm quadruped CENTAURO robot controlled with the proposed whole-body MPC for a variety of motions: a) trotting b) 5 kg object picking from the ground and c) large height reaching. Corresponding video: https://youtu.be/8XIAtw4201o
• Figure 2: The motion planning and control framework for the dual-arm quadruped CENTAURO platform. The whole-body MPC, considering the centroidal dynamics, computes optimal state and control input trajectories at 10-22 Hz which are then mapped to references for the joint impedance controllers through an inverse dynamics-based low-level reference generator module. Each MPC iteration is updated with the latest available state estimate.
• Figure 3: Cartesian velocity of a leg EE (rear right) with (red) and without (blue) the proposed cost term during a dynamic trotting motion in simulation. Penalizing cartesian leg EE velocities results in reducing impacts during leg touch-down. The shaded regions correspond to leg in contact.
• Figure 4: Snapshots of the heavy object retrieval from ground experiment described in \ref{['subsec:object_ground']}
• Figure 5: Contribution of the different torque terms (position-based feedback, velocity based feedback and feedforward torque) to the generated torque reference during free motion experiment (top) and the dynamic crawl stepping (bottom). The shaded regions denote leg in contact. Only data from the front left hip pitch joint are presented.