Learning Stable In-Grasp Manipulation in a Non-Dropping Action Space

arXiv:2606.28196v1 Announce Type: new Abstract: Traditionally, dexterous manipulation controllers are designed using analytic models constrained by strong assumptions about the hand and the objects being manipulated. Reinforcement learning (RL) has become another common approach in which skills are explored openly in an end-to-end manner but is inefficient because of unnoticeable instability and conflicts in learning objectives. This paper attempts to efficiently explore stable and accurate mani...