GuardAD: Safeguarding Autonomous Driving MLLMs via Markovian Safety Logic

2026-05-11

AI Summary

This work addresses the critical limitation of multimodal large language models (MLLMs) in autonomous driving—their lack of temporally consistent safety reasoning, which can lead to accidents in high-risk scenarios. To this end, the authors propose GuardAD, a model-agnostic safety guardrail that, for the first time, integrates higher-order Markov logical reasoning into autonomous driving safety modeling. GuardAD employs neuro-symbolic logic to formally and dynamically infer both explicit and latent risks, and instead of merely rejecting unsafe actions, it logically revises them in a reasoning-driven manner. Notably, this approach requires no modification to the underlying MLLM while enabling temporally aware safety state modeling and action optimization. Experiments demonstrate that GuardAD reduces accident rates by 32.07% and improves task performance by 6.85% across multiple benchmarks and AD-MLLMs, with effectiveness validated through both closed-loop simulation and real-world vehicle tests.

Abstract

Multimodal large language models (MLLMs) are increasingly integrated into autonomous driving (AD) systems; however, they remain vulnerable to diverse safety threats, particularly in accident-prone scenarios. Recent safeguard mechanisms have shown promise by incorporating logical constraints, yet most rely on static formulations that lack temporally grounded safety reasoning over evolving traffic interactions, resulting in limited robustness in dynamic driving environments. To address these limitations, we propose GuardAD, a model-agnostic safeguard that formulates AD safety as an evolving Markovian logical state. GuardAD introduces Neuro-Symbolic Logic Formalization, which represents safety predicates over heterogeneous traffic participants and continuously induces them via n-th order Markovian Logic Induction. This design enables the inference of emerging and latent hazards beyond single-step observations. Rather than simply vetoing unsafe actions, GuardAD performs Logic-Driven Action Revision, where inferred safety states actively guide action refinement without modifying the underlying MLLM. Extensive experiments on multiple benchmarks and AD-MLLMs demonstrate that GuardAD substantially reduces accident rates (-32.07%) while slightly improving task performance (+6.85%). Moreover, closed-loop simulation evaluations, together with physical-world vehicle studies, further validate the effectiveness and potential of GuardAD.

Problem

• Autonomous Driving
• Multimodal Large Language Models
• Safety Reasoning
• Dynamic Environments
• Markovian Logic

Innovation

• Markovian Safety Logic
• Neuro-Symbolic Logic Formalization
• Logic-Driven Action Revision
• Autonomous Driving MLLMs
• Temporal Safety Reasoning