Reproduce “Robot-Matched Soft Fingers for UMI Data Collection: Lessons from 100 Hours of Pick and Place” 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
Kobi Kelemen, Tomasz Niedziałkowski, Marcin Paszczak, Victoriano Montesinos
The authors present a handheld gripper for collecting robot-usable pick-and-place demonstrations without a robot. Its soft, angled TPU fingers match the robot gripper and visibly deform under load. The dataset contains 99.7 hours across 53 objects. On the Galaxea R1 Pro, multi-task mid-training raises success from 35% to 57% after object-specific fine-tuning on four held-out objects, but wire remains difficult.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.