Herding End-to-End Autonomous Driving via Neuro-Symbolic Safety Guards

2026-08-11

AI Summary

This work addresses the safety and interpretability limitations of end-to-end autonomous driving systems, which often violate traffic rules due to reliance on statistical data patterns and lack verifiable guarantees. The authors propose a lightweight neuro-symbolic safety guard that enforces explicit traffic regulations in real time—without retraining the underlying model or adding new learning components—by validating control commands prior to execution and substituting them with the nearest feasible safe action when necessary. Integrated seamlessly into the TransFuser v6 architecture, this mechanism ensures every intervention is traceable to a specific rule. Experiments on the Fail2Drive and Bench2Drive long-tail benchmarks demonstrate a 53% reduction in severe collisions and a 15% improvement in task success rate, all while preserving the original driving performance.

Abstract

Modern end-to-end driving agents can achieve high average performance yet still violate basic traffic rules that a human driver would never miss. The reason is structural: they learn statistical patterns rather than the physical conditions that guarantee safe driving, leaving their decision-making process opaque and safety constraints unenforced. We introduce a neuro-symbolic safety guard, a lightweight module that attaches to the final command interface of an already-trained agent. Immediately before a command reaches the vehicle, it checks the command against explicit safety rules and, only when necessary, replaces it with the nearest safe alternative. Each intervention is directly executable and traceable to the rule that triggered it, while the guard itself requires no retraining and adds no learned component. Evaluated on the long-tail benchmarks Fail2Drive and Bench2Drive using the state-of-the-art TransFuser v6 (TFv6) as a case study, the guard improves Success Rate by 15% and reduces safety-critical collisions by up to 53%, while preserving the original Driving Score.

Problem

• end-to-end autonomous driving
• safety constraints
• traffic rule violation
• decision opacity
• neuro-symbolic

Innovation

• neuro-symbolic
• safety guard
• end-to-end autonomous driving
• rule-based intervention
• collision reduction