SE(3)-Equivariant Robot Manipulation without Point Cloud Segmentation
RiEMann, an end-to-end near real-time SE(3)-equivariant robot manipulation imitation learning framework, directly predicts target object poses for manipulation without any object segmentation.