From Space to Time: Enabling Adaptive Safety with Learned Value Functions via Disturbance Recasting

2025-09-23

AI Summary

Ensuring safe control of high-dimensional autonomous systems—e.g., quadcopters—under unknown, spatially varying disturbances remains a critical challenge. Method: This paper proposes an offline learning–online adaptive safety filtering framework. Its core innovation is the first use of disturbance remapping to model spatial heterogeneity as a time-varying input, eliminating reliance on prior disturbance models. Integrating value-function-based safety filtering, offline reinforcement learning, and disturbance reparameterization, the framework enables online estimation and compensation of spatiotemporally coupled disturbances. Contribution/Results: The method is validated in both simulation and real-world experiments on a quadcopter platform. Compared to baseline approaches, it achieves significantly improved safety guarantees and trajectory tracking accuracy under complex, unknown disturbances, enabling adaptive safety assurance in unstructured environments.

Technology Category

• Intelligent Robots: State Estimation
• Natural Language Processing: Safety and Robustness
• Planning, Routing, and Scheduling: Replanning and Plan Repair

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

Abstract

The widespread deployment of autonomous systems in safety-critical environments such as urban air mobility hinges on ensuring reliable, performant, and safe operation under varying environmental conditions. One such approach, value function-based safety filters, minimally modifies a nominal controller to ensure safety. Recent advances leverage offline learned value functions to scale these safety filters to high-dimensional systems. However, these methods assume detailed priors on all possible sources of model mismatch, in the form of disturbances in the environment -- information that is rarely available in real world settings. Even in well-mapped environments like urban canyons or industrial sites, drones encounter complex, spatially-varying disturbances arising from payload-drone interaction, turbulent airflow, and other environmental factors. We introduce SPACE2TIME, which enables safe and adaptive deployment of offline-learned safety filters under unknown, spatially-varying disturbances. The key idea is to reparameterize spatial variations in disturbance as temporal variations, enabling the use of precomputed value functions during online operation. We validate SPACE2TIME on a quadcopter through extensive simulations and hardware experiments, demonstrating significant improvement over baselines.

Problem

• Ensuring autonomous system safety under varying environmental conditions
• Addressing unknown spatially-varying disturbances in safety filters
• Adapting offline-learned value functions for real-world deployment

Innovation

• Reparameterizes spatial disturbances as temporal variations
• Enables adaptive safety filters with offline-learned value functions
• Validated on quadcopter through simulations and hardware experiments