Policy enables dexterous robot hand to solve rubik's cube reliably
FINGR: Learning Dexterous Hand Control for Real-World Rubik's Cube Solving
Robotics
Summary
Solving a Rubik’s Cube with one robot hand is hard because the hand must keep the cube steady while twisting layers. The authors developed FINGR, a method that teaches the robot how each finger should move by understanding the cube’s position relative to each fingertip and predicting future moves. This method helps the robot hand turn the cube’s layers accurately and quickly. In real tests, the robot succeeded 99% of the time in individual turns and solved scrambled 2x2x2 cubes completely in about two minutes.
What this means in practice
- •For robotics engineers: Build dexterous robot hands that perform complex object manipulations like Rubik’s Cube solving under real-world conditions.
- •For automation system integrators: Integrate learned finger control policies into multi-step robotic manipulation tasks requiring stable and precise object handling.
Authors
Yutong Liang, Quanquan Peng, Matthew Kim, Xiaolong Wang
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