The CosmoH2G dataset contains 6,189 episodes across 1,254 unique objects, specifically designed to capture complex spatial movements like rotations and flips. It provides a scalable pipeline for transferring human hand demonstrations to robotic grippers via fine-grained hand-pose motions.
HOW THIS AFFECTS YOU
●
builderYou can train robotic grippers to perform intricate spatial manipulations using human hand data.
●
researcherThis dataset addresses the limitations of existing planar-task datasets in robot learning.