Cognitive Dual-Process Planning for Autonomous Driving with Structured Scene Knowledge and Verifiable Reasoning-Action Consistency
作者: Zhongyao Yang, Haoyu Li, Yu Yan, Zhuangxuan Yu, Jiangfeng Nan, Jinrui Nan
分类: cs.RO, cs.CV
发布日期: 2026-07-21
备注: 12 pages, 7 figures. Zhongyao Yang and Haoyu Li contributed equally to this work
💡 一句话要点
提出认知双过程规划框架以解决自主驾驶中的场景理解问题
🎯 匹配领域: 支柱三:空间感知与语义 (Perception & Semantics) 支柱九:具身大模型 (Embodied Foundation Models)
关键词: 自主驾驶 高层规划 视觉-语言模型 结构化推理 场景理解 认知框架 策略优化
📋 核心要点
- 现有方法在自主驾驶的高层规划中面临场景理解不准确和推理不一致的问题,限制了其应用效果。
- 本文提出了一种认知双过程规划框架,通过结构化思维链(S-CoT)表示场景知识,结合自动化数据引擎生成监督信息。
- 实验结果显示,生成的注释在91.8%的思维链准确率和98.5%的逻辑一致性得分上表现优异,规划准确率达到80.14%。
📝 摘要(中文)
高层规划在自主驾驶中是一项知识密集型的工程决策任务,要求准确的场景理解、及时的推理和内部一致的行动选择。尽管视觉-语言模型(VLMs)能够使中间推理显性化,但在实际应用中受到结构化监督成本高、常规场景中不必要的推理以及生成的推理与驾驶行动之间可能存在不一致性的限制。本文提出了一种认知双过程规划框架,通过机器可解析的结构化思维链(S-CoT)模式表示与规划相关的场景知识。自动化数据引擎整合了感知基础模型、关键路径过滤和专家VLM,生成S-CoT监督而无需手动注释个别推理。在多层视觉编码特征的基础上,轻量级视觉仲裁者在语言解码之前评估场景复杂性,并将每个输入路由到快速元行动预测或慢速结构化推理。对于慢路径输出,确定性基于规则的验证器检查解析的S-CoT字段是否与最终元行动一致,并为群体相对策略优化(GRPO)提供可验证的奖励。
🔬 方法详解
问题定义:本文旨在解决自主驾驶中高层规划的场景理解和推理一致性问题。现有方法在结构化监督和推理效率上存在不足,导致决策质量下降。
核心思路:提出认知双过程规划框架,通过机器可解析的结构化思维链(S-CoT)来表示与规划相关的场景知识,结合自动化数据引擎生成监督信息,提升推理效率和一致性。
技术框架:整体架构包括感知基础模型、关键路径过滤、专家VLM和轻量级视觉仲裁者。视觉仲裁者根据场景复杂性将输入分为快速元行动预测和慢速结构化推理两条路径。
关键创新:最重要的创新在于通过结构化思维链(S-CoT)实现了推理过程的显性化和可验证性,解决了传统方法中推理与行动不一致的问题。
关键设计:在设计中,使用了确定性规则验证器来检查S-CoT字段与最终元行动的一致性,并为群体相对策略优化(GRPO)提供可验证的奖励,确保推理过程的可靠性。
🖼️ 关键图片
📊 实验亮点
实验结果显示,生成的注释在195个场景的手动审核中达到了91.8%的思维链准确率和98.5%的逻辑一致性得分。在574个手动验证的NAVSIM测试样本中,规划准确率为80.14%,逻辑一致性得分为97.20%,同时平均延迟减少了17.39%。
🎯 应用场景
该研究的潜在应用领域包括自动驾驶系统、智能交通管理和机器人导航等。通过提升场景理解和推理一致性,能够显著提高自主驾驶的安全性和效率,推动智能交通技术的发展。未来,该框架有望在更复杂的动态环境中应用,进一步增强自主系统的决策能力。
📄 摘要(原文)
High-level planning for autonomous driving is a knowledge-intensive engineering decision task that requires accurate scene understanding, timely inference, and internally consistent action selection. Vision-language models (VLMs) can make intermediate reasoning explicit, but their use in deployed planners is constrained by costly structured supervision, unnecessary reasoning in routine scenes, and possible inconsistencies between generated rationales and driving actions. We present a cognitive dual-process planning framework that represents planning-relevant scene knowledge in a machine-parsable structured chain-of-thought (S-CoT) schema. An automated data engine integrates perception foundation models, critical-path filtering, and an expert VLM to generate S-CoT supervision without manual annotation of individual rationales. A lightweight visual Arbiter estimates scene complexity from multilevel vision-encoder features before language decoding and routes each input to either fast meta-action prediction or slow structured reasoning. For slow-path outputs, a deterministic rule-based validator checks whether the parsed S-CoT fields are consistent with the final meta-action and provides verifiable rewards for Group Relative Policy Optimization (GRPO). In a 195-scene manual audit, the generated annotations achieve 91.8\% CoT accuracy and a 98.5\% Logical Consistency Score (LCS). On 574 manually verified NAVSIM test samples, the planner achieves 80.14\% planning accuracy and 97.20\% LCS while reducing average latency by 17.39\% relative to applying slow reasoning to every scene. Evaluation on external long-tail subsets further identifies conditions under which routing and planning performance degrade. Together, these results show how explicit scene knowledge can be operationalized through adaptive reasoning and rule-based verification to support high-level VLM planning decisions.