DexFLEX improves robot hand control by making contact decisions
DexFLEX: Contact-Aware Foundation Controller for Command-Guided Dexterity
RoboticsArtificial IntelligenceMachine Learning
Summary
Robot hands often receive instructions on how to move, but these instructions don’t say much about when or how to touch or hold things securely. The authors introduce DexFLEX, a controller that helps robot hands decide how to move their fingers while making good contact with objects. Instead of blindly copying finger movements, DexFLEX predicts the effects of movements and chooses the best motions to keep stable contact and follow the command. This approach made teleoperated robot hand tasks much more successful and improved learning-based control.
What this means in practice
- •For robotics engineers: Enhance teleoperated robot hand performance by integrating contact-aware control to increase grasping and manipulation success.
- •For robotics software developers: Improve robustness of learned robot hand policies by correcting fingertip motion commands into stable, contact-consistent joint actions without retraining.
- •For prosthetics developers: Use contact-aware control for dexterous prosthetic hands to better interpret user intents while maintaining stable grip on objects.$Commercial implications: Provides a foundation for more reliable prosthetic hand controllers that better translate user commands into successful object interaction.
Authors
Weikang Wan, Zeyu Lai, Jianyu Hou, Yuchen Zhou, Jiawei Fu, Henrik I Christensen, Hao Su
Abstract
Dexterous robot hands can receive useful motion intent from teleoperation or learned policies, but successful execution also depends on contact decisions that these commands rarely specify. We introduce DexFLEX, a contact-aware foundation controller that turns upstream fingertip-motion drafts into contact-consistent joint commands. Instead of treating a draft as a trajectory to copy, DexFLEX treats it as evidence about intent: from the current tactile-proprioceptive state, it proposes feasible short-horizon motion chunks, predicts their contact consequences, and selects the candidate that best follows the command while preserving future contact stability. At inference time, a simple Draft-Dream-Select procedure combines pure-prior proposals for recovery with draft-seeded proposals for responsiveness, then decodes the selected motion into executable joint-space targets. Across degraded-command simulation, real-world shared control, and visuomotor policy learning, the same trained controller improves robustness without retraining, increasing real-world teleoperation success from 29.2% to 78.3% and reaching 46.7% policy-learning success, 3.5 times direct joint-action prediction, and 1.56 times prior-only correction. Videos are available at dex-flex.github.io.