Reinforcement learning fine-tuning on top of large, pretrained Vision-Language-Action (VLA) models offers promise for highly reliable robot deployment. Howev…
机构:Stanford
来源:arXiv 2609.18207 | AI4Papers 论文推荐平台