LaST-VLA: 在自动驾驶的视觉-语言-动作中思考于潜在的空间-时间空间
LaST-VLA: Thinking in Latent Spatio-Temporal Space for Vision-Language-Action in Autonomous Driving
AI总结：
LaST-VLA通过潜在空间-时间推理框架，结合双特征对齐机制和渐进式SFT训练策略，提升自动驾驶中的视觉-语言-动作处理能力，实现更安全的物理基础推理。
摘要：
尽管视觉-语言-动作（VLA）模型通过统一感知和规划革新了自动驾驶，但其对显式文本链式思维（CoT）的依赖导致了语义-感知解耦和感知-符号冲突。最近的转变倾向潜在推理试图通过在连续隐藏空间中思考来绕过这些瓶颈。然而，没有显式的中间约束，标准的潜在CoT通常作为物理无关的表示。为了解决这个问题，我们提出了潜在空间-时间VLA（LaST-VLA），一种将推理范式从离散符号处理转向物理基础的潜在空间-时间CoT的框架。通过实现双特征对齐机制，我们从3D基础模型中蒸馏几何约束，并从世界模型中提取动态前瞻性，直接输入潜在空间。结合一种渐进式SFT训练策略，从特征对齐过渡到轨迹生成，并通过强化学习与组相对策略优化（GRPO）进行精炼，以确保安全性和规则遵守。在NAVSIM v1（91.3 PDMS）和NAVSIM v2（87.1 EPDMS）上设定了新的记录，同时在SURDS和NuDynamics基准上表现出色的空间-时间推理能力。
英文摘要：
While Vision-Language-Action (VLA) models have revolutionized autonomous driving by unifying perception and planning, their reliance on explicit textual Chain-of-Thought (CoT) leads to semantic-perceptual decoupling and perceptual-symbolic conflicts. Recent shifts toward latent reasoning attempt to bypass these bottlenecks by thinking in continuous hidden space. However, without explicit intermediate constraints, standard latent CoT often operates as a physics-agnostic representation. To address this, we propose the Latent Spatio-Temporal VLA (LaST-VLA), a framework shifting the reasoning paradigm from discrete symbolic processing into a physically grounded Latent Spatio-Temporal CoT. By implementing a dual-feature alignment mechanism, we distill geometric constraints from 3D foundation models and dynamic foresight from world models directly into the latent space. Coupled with a progressive SFT training strategy that transitions from feature alignment to trajectory generation, and refined via Reinforcement Learning with Group Relative Policy Optimization (GRPO) to ensure safety and rule compliance. \method~setting a new record on NAVSIM v1 (91.3 PDMS) and NAVSIM v2 (87.1 EPDMS), while excelling in spatial-temporal reasoning on SURDS and NuDynamics benchmarks.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
