CHIP: Adaptive Compliance for Humanoid Control through Hindsight Perturbation

2025-12-16

AI Summary

Humanoid robots face a fundamental challenge in force-intensive tasks—such as cart pushing, surface wiping, and door opening—where simultaneous high-fidelity end-effector stiffness regulation and dynamic motion tracking are difficult to achieve. To address this, we propose CHIP, a plug-and-play control module. Its core innovation is the first-ever online stiffness adaptation mechanism based on hindsight perturbation, requiring neither data augmentation nor reward redesign. CHIP unifies end-to-end motion tracking with model-free compliant parameter estimation, enabling a single generic controller to generalize across diverse physical interaction tasks. We validate CHIP across multi-robot collaboration, box transport, wiping, and door-opening scenarios. Results demonstrate substantial improvements in the joint performance of force control accuracy and motion agility, outperforming prior approaches in both task success rate and trajectory fidelity under contact-rich dynamics.

Technology Category

• Intelligent Robots: Manipulation
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Computer Vision: Motion & Tracking

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Economics, Online Markets and Human Computation: Research challenges in human and human-AI computation
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Recent progress in humanoid robots has unlocked agile locomotion skills, including backflipping, running, and crawling. Yet it remains challenging for a humanoid robot to perform forceful manipulation tasks such as moving objects, wiping, and pushing a cart. We propose adaptive Compliance Humanoid control through hIsight Perturbation (CHIP), a plug-and-play module that enables controllable end-effector stiffness while preserving agile tracking of dynamic reference motions. CHIP is easy to implement and requires neither data augmentation nor additional reward tuning. We show that a generalist motion-tracking controller trained with CHIP can perform a diverse set of forceful manipulation tasks that require different end-effector compliance, such as multi-robot collaboration, wiping, box delivery, and door opening.

Problem

• Enables controllable end-effector stiffness for humanoid robots
• Performs forceful manipulation tasks without data augmentation
• Adapts compliance for diverse tasks like wiping and door opening

Innovation

• Adaptive compliance control for humanoid robots
• Plug-and-play module without data augmentation
• Enables forceful manipulation while preserving agile motion