Recent multimodal large language models (MLLMs) can perceive a traffic scene, reason about it, and propose a driving action within a single model. However, e…
机构:华盛顿大学
来源:arXiv 2610.32069 | AI4Papers 论文推荐平台