Reproduce “Uni-PNP: Unified Prehensile and Non-prehensile Policy Learning for Generalizable Object Pose Reaching” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Wenxuan Zhu, Yuze Wang, Jiangran Lyu, Jiayi Chen, Yixin Zheng, Mi Yan, Li Yi, Zhizheng Zhang and 1 more

Uni-PNP coordinates non-prehensile repositioning and grasping to move tabletop objects toward target poses. It pretrains an encoder to predict object motion from gripper actions, then trains a policy from graspable to harder objects. In simulation, it reaches 57.91% success on unseen objects with Panda, versus 40.12% for UniCORN; evidence is limited to tabletop pose reaching with two-finger grippers.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.