Faster Visuomotor Policy Learning on Action Manifolds via Riemannian MeanFlow figure
AlphaXiv 中文概览(可滚动查看)