Problem
Research questions and friction points this paper is trying to address.
dexterous hand
Rubik's Cube
contact-rich manipulation
real-world control
FINGR: Learning Dexterous Hand Control for Real-World Rubik's Cube Solving
🤖 AI Summary
This study addresses the challenge of coordinating support and rotation actions under continuous contact during one-handed dexterous manipulation of a Rubik's cube. To this end, it proposes FINGR, a policy that leverages a shared encoder to fuse finger-relative geometric features with future interaction predictions for action generation. The framework incorporates multi-timescale future supervision signals—encompassing contact force, progress, and displacement—and employs an index-agnostic point cloud aggregation mechanism to enhance generalization. Furthermore, it integrates a flow-based policy with a regrasping assistance system. In real-world experiments, the proposed approach achieves a 99.0% success rate for single rotations and fully solves a 2×2×2 Rubik's cube in an average of 137 seconds.
📝 Abstract
Manipulating a Rubik's Cube with a single dexterous hand is a challenging test of sustained, contact-rich control: the hand must execute successive layer turns while keeping the cube secure. Each turn requires some fingers to support the cube while others push a moving layer, release contact, and reset for the next move. To learn this coordination, we introduce FINGR (Future-supervised Interaction Network with Geometric Representations), a policy that combines finger-relative geometry with future interaction prediction. A shared point encoder expresses the cube relative to each fingertip and aggregates its points without depending on cubie indexing. Learned future tokens share the observation encoder and receive supervision for contact-force changes, layer-turn progress, and finger joint displacement at multiple time scales. The resulting representation conditions a flow policy that directly generates finger actions. On a real dexterous hand, our policy achieves 99.0% success over 300 turn attempts, compared with 79.7% for the base flow policy. Integrated with grasping and table-assisted regrasping, the policy solves all ten scrambled $2\times2\times2$ cubes in a mean complete-system time of approximately 137 seconds. The project website is available at https://www.lyt0112.com/projects/FINGR
Innovation
Methods, ideas, or system contributions that make the work stand out.
Dexterous Hand Control
Future Interaction Prediction
Geometric Representations
Flow Policy
Rubik's Cube Solving
🔎 Similar Papers
No similar papers found.
