Reproduce “Direct Action-Head Injection of A Grounded 3D Point Unlocks Spatial and Task Generalization” 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
Shiang-Feng Tsai, Jin-Cheng Jhang, Yen-Ling Tai, Jia-Hong Lai, Shih-Yun Wong, Kang-Tung Hsu, Yi-Ting Chen, Min Sun
The method conditions GR00T-N1.6’s action head on a 3D gripper-to-target displacement derived from a grounded image point. It lifts the point using depth and calibrated camera parameters, subtracts the gripper position, and injects the encoded displacement through adaptive layer normalization. On LIBERO-PRO, average success rises from 31.2% to 77.5% under task perturbation and from 28.1% to 60.2% under position perturbation, for tested pick-and-place shifts with adequate grounding.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.