Vision-Language-Action (VLA) models have shown promising progress in language-conditioned robotic manipulation. However, existing datasets and benchmarks mai…
机构:浙大
来源:arXiv 2609.05324 | AI4Papers 论文推荐平台