Reproduce “PointPilot: Unified Robotic Manipulation via Object-Centric 3D Point Flow Prediction” 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
Yihan Fang, Yuxin Tian, Ruixiang Wang, Zhanqi Zhang, Xing Hu, Zhen Yang, Huayi Zhou, Yongyi Su and 2 more
PointPilot uses task-relevant 3D point flows to support manipulation across robot embodiments. It predicts future motion for gripper or object points, assigns flows to grippers, and maps them to end-effector poses through a geometric adapter. In a seven-task RoboTwin transfer evaluation, average success fell from 86.0% on ALOHA to 71.0% on Piper, on selected feasible paired scenes with robot-specific setup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.