Unified Humanoid Fall-Safety Policy from a Few Demonstrations

2025-11-10

AI Summary

Humanoid robots are prone to falling under dynamic imbalance, and existing approaches address fall prevention, impact mitigation, or recovery in isolation—lacking a holistic, autonomous safety strategy across the entire fall cycle. This paper introduces the first end-to-end control framework that unifies modeling of the complete fall process (prevention → mitigation → recovery). It integrates sparse human demonstrations, reinforcement learning, and a diffusion-model-based adaptive memory mechanism to enable policy generalization and efficient sim-to-real transfer. Evaluated on the Unitree G1 platform, the method significantly reduces ground-impact forces and achieves stable, rapid autonomous recovery under diverse disturbances. It demonstrates superior robustness and empirical transfer capability compared to state-of-the-art modular approaches. The framework provides a scalable, safety-critical autonomous solution for real-world imbalance recovery in humanoid robotics.

Technology Category

• Intelligent Robots: Manipulation
• Natural Language Processing: Safety and Robustness
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Falling is an inherent risk of humanoid mobility. Maintaining stability is thus a primary safety focus in robot control and learning, yet no existing approach fully averts loss of balance. When instability does occur, prior work addresses only isolated aspects of falling: avoiding falls, choreographing a controlled descent, or standing up afterward. Consequently, humanoid robots lack integrated strategies for impact mitigation and prompt recovery when real falls defy these scripts. We aim to go beyond keeping balance to make the entire fall-and-recovery process safe and autonomous: prevent falls when possible, reduce impact when unavoidable, and stand up when fallen. By fusing sparse human demonstrations with reinforcement learning and an adaptive diffusion-based memory of safe reactions, we learn adaptive whole-body behaviors that unify fall prevention, impact mitigation, and rapid recovery in one policy. Experiments in simulation and on a Unitree G1 demonstrate robust sim-to-real transfer, lower impact forces, and consistently fast recovery across diverse disturbances, pointing towards safer, more resilient humanoids in real environments. Videos are available at https://firm2025.github.io/.

Problem

• Develops unified policy for humanoid fall prevention and recovery
• Integrates impact mitigation with rapid standing after falls
• Combines human demonstrations with reinforcement learning for safety

Innovation

• Fuses human demonstrations with reinforcement learning
• Uses adaptive diffusion-based memory for safe reactions
• Unifies fall prevention, impact mitigation, and recovery