Textual reasoning before acting has improved vision-language-action (VLA) models, but at the cost of latency: every action conditioned on the reasoning text …
机构:NVIDIA
来源:arXiv 2610.17540 | AI4Papers 论文推荐平台