Event
MAP Interface Unifies Spatiotemporal References as an Action Repair Problem
1 reports · 1 sources
Yanbang Li et al. propose the MAP (Masked Action Space) interface, which unifies spatiotemporal references such as keypoints, flows, orientations, and key poses as partial constraints in the same temporal Cartesian action space, and models controllable robot manipulation as an action repair problem. On ManiSkill and LIBERO-10, the average task success rates of this strategy are 92.67% and 64.67%, respectively; the study also adopts a training-free RePaint variant that iteratively repairs actions during the denoising process to reduce control error. Real-robot experiments show that the average position error and rotation error of spatiotemporal control are 44.3 mm and 7°, respectively.
Report timeline
- Robotic Spatiotemporal Editing: A Unified and Controllable Action Generation Interface via Masked Action Reconstruction
The paper proposes the MAP (Masked Action Space) interface, which unifies spatiotemporal references such as keypoints, flows, orientations, and key poses as partial constraints in the same temporal Cartesian action space, and models controllable manipulation as an action inpainting problem.
arXiv cs.RO ·