Bracing for Impact: Robust Humanoid Push Recovery and Locomotion with Reduced Order Models

TL;DR

This work tackles robust push recovery for humanoid locomotion in human environments by enabling wall bracing via the robot arms. It fuses a Single Rigid Body MPC with Hybrid Linear Inverted Pendulum dynamics to form a unified controller that detects pushes, adapts contact forces, and modifies stepping while walking. The approach demonstrates superior perturbation rejection and tracking in high fidelity MuJoCo simulations, recovering from pushes up to $100\,\text{N}$ within $0.2\text{s}$ and maintaining commanded speeds up to $0.5\,\text{m/s}$. The results indicate significant gains in safe operating envelopes and robustness to wall geometry and multi direction pushes, supporting real time environment aware push recovery for humanoids. The framework paves the way for hardware deployment and extension to diverse humanoid platforms.

Abstract

Push recovery during locomotion will facilitate the deployment of humanoid robots in human-centered environments. In this paper, we present a unified framework for walking control and push recovery for humanoid robots, leveraging the arms for push recovery while dynamically walking. The key innovation is to use the environment, such as walls, to facilitate push recovery by combining Single Rigid Body model predictive control (SRB-MPC) with Hybrid Linear Inverted Pendulum (HLIP) dynamics to enable robust locomotion, push detection, and recovery by utilizing the robot's arms to brace against such walls and dynamically adjusting the desired contact forces and stepping patterns. Extensive simulation results on a humanoid robot demonstrate improved perturbation rejection and tracking performance compared to HLIP alone, with the robot able to recover from pushes up to 100N for 0.2s while walking at commanded speeds up to 0.5m/s. Robustness is further validated in scenarios with angled walls and multi-directional pushes.

Figures (5)

• Figure 1: The proposed SRB-MPC-HLIP controller successfully utilizes the arms of the humanoid robot and adjusts its support polygon dynamically to brace the robot against walls in the event of a sudden push, while the HLIP controller could not account for such a sudden perturbation and fails to regain stable locomotion.
• Figure 2: Framework for the SRB-MPC-HLIP controller. The robot runs state estimation and wall detection based on the sensor data from the environment as illustrated by the red lines in the sensing & state estimation module, and the SRB-MPC and HLIP modules runs in tandem to reject large perturbations utilizing the environment by activating the recovery mode when necessary.
• Figure 3: Tracking performance of the pure HLIP controller and the SRB-MPC-HLIP controller. The proposed controller can maintain adequate command tracking and better stabilization of the torso orientation. The snapshot illustrates the robot accelerating to a sagittal velocity of $0.5$ m/s.
• Figure 4: (a) Worst-case all-direction safeset of both controllers when pushed while stepping in place with no environmental assistance. The proposed controller has a $225\%$ larger safeset (b) Safeset of the HLIP controller covering $16.8\%$ of total points tested. (c) Safeset of the SRB-MPC-HLIP controller covering $70.6\%$ of total points tested, a $420\%$ increase over the HLIP controller. Snapshots of one easy scenario ($40$N force at $0.1$m above the CoM) and one hard scenario ($80$N force at $0.5$m above the CoM) are also illustrated. It can be seen that the proposed SRB-MPC-HLIP controller maintains better torso stability. The orange arrows in the snapshots show the hand contacts.
• Figure 5: Snapshots from simulated experiments,, the robot is pushed with $[-20\text{N},30\text{N}]$ from the right at $2s$ and $[10\text{N}, -40\text{N}]$ from the left at $12$s. The wall is slanted inwards at 5 degrees. The robot was successful in utilizing its arms to brace against the walls to recover stable walking.