Learning Safe-Stoppability Monitors for Humanoid Robots

2026-03-23

AI Summary

This work addresses the challenge of emergency stopping in humanoid robots, where conventional mechanisms often lead to instability or damage due to abrupt power interruption. The authors formulate the emergency stop problem as a policy-dependent safe-stoppability issue and introduce PRISM, a novel framework that employs a neural network to learn a state-level stoppability predictor. By integrating importance sampling for targeted exploration and iteratively refining the decision boundary in simulation, PRISM significantly improves data efficiency and reduces false-positive safety predictions. The method accurately captures the stoppability boundary and successfully transfers from simulation to real-world deployment. A pre-trained monitor is implemented on a physical humanoid robot, enabling proactive safety monitoring and facilitating scalable certification of fail-safe behaviors.

Technology Category

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

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Security and Privacy: Security and privacy of machine learning and AI applications

Abstract

Emergency stop (E-stop) mechanisms are the de facto standard for robot safety. However, for humanoid robots, abruptly cutting power can itself cause catastrophic failures; instead, an emergency stop must execute a predefined fallback controller that preserves balance and drives the robot toward a minimum-risk condition. This raises a critical question: from which states can a humanoid robot safely execute such a stop? In this work, we formalize emergency stopping for humanoids as a policy-dependent safe-stoppability problem and use data-driven approaches to characterize the safe-stoppable envelope. We introduce PRISM (Proactive Refinement of Importance-sampled Stoppability Monitor), a simulation-driven framework that learns a neural predictor for state-level stoppability. PRISM iteratively refines the decision boundary using importance sampling, enabling targeted exploration of rare but safety-critical states. This targeted exploration significantly improves data efficiency while reducing false-safe predictions under a fixed simulation budget. We further demonstrate sim-to-real transfer by deploying the pretrained monitor on a real humanoid platform. Results show that modeling safety as policy-dependent stoppability enables proactive safety monitoring and supports scalable certification of fail-safe behaviors for humanoid robots.

Problem

• safe-stoppability
• humanoid robots
• emergency stop
• fallback controller
• safety certification

Innovation

• safe-stoppability
• humanoid robots
• importance sampling
• simulation-to-real transfer
• neural safety monitor