Learning to Navigate Under Imperfect Perception: Conformalised Segmentation for Safe Reinforcement Learning

2025-10-21

AI Summary

In safety-critical navigation, imperfect perception introduces significant risk in hazard avoidance—existing approaches either assume perfect hazard detection or lack rigorous, finite-sample safety guarantees. Method: We propose COPPOL, the first framework to integrate distribution-free, finite-sample safety guarantees directly into the perception module. It unifies conformal prediction with semantic segmentation to produce a calibrated hazard map and a risk-aware cost field with provable upper bounds on missed detections; this field then guides uncertainty-aware reinforcement learning for planning. Results: Evaluated on two satellite remote sensing benchmarks, COPPOL achieves up to 6× improvement in hazardous region coverage and reduces hazard violation rates by ~50%, while maintaining robustness under distributional shift. Its core contribution is a verifiable, end-to-end propagation of perceptual uncertainty into decision-time safety boundaries.

Technology Category

• Natural Language Processing: Safety and Robustness
• Reasoning under Uncertainty: Uncertainty Representations
• Planning, Routing, and Scheduling: Planning under Uncertainty

Application Category

• Security and Privacy: Data transparency and provenance
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Reliable navigation in safety-critical environments requires both accurate hazard perception and principled uncertainty handling to strengthen downstream safety handling. Despite the effectiveness of existing approaches, they assume perfect hazard detection capabilities, while uncertainty-aware perception approaches lack finite-sample guarantees. We present COPPOL, a conformal-driven perception-to-policy learning approach that integrates distribution-free, finite-sample safety guarantees into semantic segmentation, yielding calibrated hazard maps with rigorous bounds for missed detections. These maps induce risk-aware cost fields for downstream RL planning. Across two satellite-derived benchmarks, COPPOL increases hazard coverage (up to 6x) compared to comparative baselines, achieving near-complete detection of unsafe regions while reducing hazardous violations during navigation (up to approx 50%). More importantly, our approach remains robust to distributional shift, preserving both safety and efficiency.

Problem

• Addresses unreliable hazard perception in safety-critical navigation tasks
• Provides finite-sample safety guarantees for semantic segmentation uncertainty
• Enables risk-aware reinforcement learning with calibrated hazard detection

Innovation

• Conformal-driven perception-to-policy learning approach
• Integrates safety guarantees into semantic segmentation
• Generates risk-aware cost fields for RL planning