REDOUBT: Duo Safety Validation for Autonomous Vehicle Motion Planning

Safety validation, which assesses the safety of an autonomous system's motion planning decisions, is critical for the safe deployment of autonomous vehicles. Existing input validation techniques from other machine learning domains, such as image classification, face unique challenges in motion planning due to its contextual properties, including complex inputs and one-to-many mapping. Furthermore, current output validation methods in autonomous driving primarily focus on openloop trajectory prediction, which is ill-suited for the closed-loop nature of motion planning. We introduce REDOUBT, the first systematic safety validation framework for autonomous vehicle motion planning that employs a duo mechanism, simultaneously inspecting input distributions and output uncertainty. REDOUBT identifies previously overlooked unsafe modes arising from the interplay of In-Distribution/Out-of-Distribution (OOD) scenarios and certain/uncertain planning decisions. We develop specialized solutions for both OOD detection via latent flow matching and decision uncertainty estimation via an energy-based approach. Our extensive experiments demonstrate that both modules outperform existing approaches, under both open-loop and closed-loop evaluation settings. Our codes are available at: https://github.com/sgNicola/Redoubt.

Safety validation for motion planning, which assesses whether a vehicle's decisions are safe in a given scenario, is thus critical. When an unsafe decision is detected, the validation module can: (1) immediately override the decision and initiate fallback measures (e.g., manual takeover [1]) to ensure short-term safety, and (2) flag the scenario as "challenging" to prioritize improvement in subsequent continual learning iterations [37,41]. Safety validation could reduce or prevent tragedies like the March 29, 2025, Xiaomi SU7 incident, which resulted in three fatalities [12]. Nevertheless, such a systematic safety validation framework is currently lacking-a gap our this work seeks to address.

Safety validation can potentially be approached via two main angles: input inspection and output inspection. In the first category, input inspection, we draw inspiration from Out-of-Distribution 39th Conference on Neural Information Processing Systems (NeurIPS 2025).