1 paper
Xiangyu Zhu, Jin Xu, Yue Guo +6
Video generation models (VGMs) offer strong spatiotemporal priors for embodied observation--action modeling. However, joint-space action vectors lack explicit image-space structure…