ATOM-CBF: Adaptive Safe Perception-Based Control under Out-of-Distribution Measurements

2025-11-11

AI Summary

Learning-based perception modules pose safety risks under out-of-distribution (OoD) inputs due to epistemic uncertainty. Existing safety frameworks typically require ground-truth labels or prior knowledge of distribution shifts—often unavailable in real-world deployment. Method: We propose a label-free, distribution-agnostic adaptive safety control framework that explicitly models OoD perception uncertainty and embeds it into Control Barrier Functions (CBFs). Our approach jointly optimizes data-driven perception models with a safety layer, integrating online adaptive error-bound estimation and a safety-filtering mechanism to enable tunable conservatism and real-time guarantees. Contribution/Results: To our knowledge, this is the first work to incorporate explicit OoD uncertainty modeling directly into CBF synthesis. Evaluated on F1Tenth LiDAR navigation and quadrupedal robot RGB-vision control tasks, our method significantly improves safety and robustness under OoD conditions. It establishes a verifiable, deployable safety paradigm for autonomous systems relying on learned perception.

Technology Category

• Natural Language Processing: Safety and Robustness
• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Uncertainty Representations

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work

Abstract

Ensuring the safety of real-world systems is challenging, especially when they rely on learned perception modules to infer the system state from high-dimensional sensor data. These perception modules are vulnerable to epistemic uncertainty, often failing when encountering out-of-distribution (OoD) measurements not seen during training. To address this gap, we introduce ATOM-CBF (Adaptive-To-OoD-Measurement Control Barrier Function), a novel safe control framework that explicitly computes and adapts to the epistemic uncertainty from OoD measurements, without the need for ground-truth labels or information on distribution shifts. Our approach features two key components: (1) an OoD-aware adaptive perception error margin and (2) a safety filter that integrates this adaptive error margin, enabling the filter to adjust its conservatism in real-time. We provide empirical validation in simulations, demonstrating that ATOM-CBF maintains safety for an F1Tenth vehicle with LiDAR scans and a quadruped robot with RGB images.

Problem

• Addresses safety challenges in systems using learned perception modules
• Mitigates epistemic uncertainty from out-of-distribution sensor measurements
• Provides adaptive safety control without requiring ground-truth labels

Innovation

• Adaptive safe control for out-of-distribution measurements
• OoD-aware perception error margin for real-time adaptation
• Safety filter integrating adaptive error margin