Language-Structured Relational Q-Learning for Threat-Aware Control in Safety-Critical Driving

What happened

Analysis and impact

导读摘要

正文

这篇题为《Language-Structured Relational Q-Learning for Threat-Aware Control in Safety-Critical Driving》的论文于 2026 年 8 月 13 日发布在 arXiv 上，归属于 cs.CV 领域。根据摘要，研究团队指出，基于自然语言的场景生成提供了一种直观的方式来描述稀有且复杂的驾驶交互场景，但用这类生成数据训练模型是否有效，此前仍存在不确定性。该工作试图将语言结构化信息引入关系型 Q 学习框架，以提升自动驾驶系统在安全关键场景下的威胁感知与控制能力。

从方法层面看，论文的核心思路是将自然语言描述的场景语义作为结构化先验，嵌入到关系型强化学习的训练过程中。关系型 Q 学习本身擅长处理多智能体之间的交互关系，而自然语言生成场景则能够覆盖那些在真实数据中极少出现、却对安全性至关重要的边缘案例。两者结合的目标，是让模型在面对罕见但危险的驾驶情境时，能够更准确地识别威胁并做出相应决策。摘要中关于具体实验设置、数据集规模与性能指标的细节，原文未提供完整信息。

值得注意的是，同日 arXiv 上还发布了多篇与强化学习及智能体训练相关的工作。例如一篇关于「Self-Evolving Embodied Agents via Skill-Harness Evolution」的论文（cs.CL 领域）探讨了围绕基础模型构建的具身智能体如何通过技能演化实现自我进化；另一篇来自 cs.AI 领域的论文则聚焦于利用强化学习控制 LLM 训练过程中的 GPU 功耗，从单卡扩展到整个集群。这些工作共同反映出当前强化学习在自动驾驶、具身智能与基础设施效率等多个方向上的活跃态势。

就自动驾驶安全研究而言，稀有场景的覆盖一直是行业痛点。真实路测数据中，严重事故场景天然稀缺，而模拟器生成的场景又往往与真实分布存在偏差。语言驱动的场景生成提供了一种更灵活的补充路径——研究者可以用自然语言描述「夜间雨天、前方车辆突然变道、对向有行人横穿」等组合条件，系统据此生成对应的训练场景。这篇论文的贡献在于尝试回答一个关键问题：这类语言生成场景是否真的能有效提升模型在安全关键任务上的表现，而不仅仅是扩充了数据量。

论文配图未提供

References

Original source text

arXiv:2608.11498v1 Announce Type: new Abstract: Natural-language-based scenario generation offers an intuitive means of describing rare and complex driving interactions, yet it is still uncertain whether training with language-structured data leads to truly adaptive control policies. We propose Language-Structured Relational Q-Learning, instantiated through an Ego-Centric Relational Q-Network (ERQ-Net), which jointly learns inter-vehicle relevance and action values from dynamic traffic graphs. Language descriptions define surrounding-vehicle behaviours during training, while prompts and semantic actor roles are hidden from the policy. ERQ-Net must therefore infer threat relevance solely from observable kinematics and interactions. Across 2,500 safety-critical scenarios, language-structured training improves test success from 49-52% to 55-58% and increases adversary-focused attention from 1.2x to 2.1x, demonstrating emergent threat awareness. However, this representational gain does not consistently translate into adaptive control: trained policies perform similarly to the best constant action, while a portfolio of simple policies solves 76% of scenarios. We formalise this discrepancy as a recognition-control gap and show that reward reweighting and margin shaping do not eliminate the resulting policy collapse. Evaluations of realism, criticality, semantic accuracy, and transfer of state-interface representations to CARLA further highlight both the strengths and the constraints of language-structured relational policy learning in safety-critical driving scenarios.