Overview
This post shares the paper Aero Hand Open: A Simulation-Ready Tendon-Driven Hand for Dexterous Manipulation (arXiv:2608.28578).
- Field: Machine Learning / Robotics
- Authors: Nan Wang, Mohit Yadav, Jonathan Wulff, Aidan Rosenbaum, Kezhou Chen, Yuvan Sharma, Xu Dong, Yiwei Tao
- Posted: 2026-08-28
- Tendon-driven hands are anthropomorphic, and moving actuators off the joints makes a hand of this capability affordable to build. Routing force through cables removes the requirement that a motor fit inside the joint it drives, so smaller and cheaper motors suffice; one motor can also drive several joints through a single cable, so fewer motors are needed.
- Tendon-driven hands are harder to learn on than direct-drive hands: the underactuated transmission that produces the cost savings is difficult to represent in a simulator, and joints driven by one cable cannot be commanded independently.
- Aero Hand Open is a tendon-driven anthropomorphic hand released simulation-ready, shipping with three components:
- A simulation model that reproduces the cable transmission itself.
- An identified actuation mapping that connects the simulation model to motor commands bidirectionally, including the three-way coupling of the thumb.
- A reinforcement learning package for training policies on the hand.
- Combined, these allow policies to be trained entirely in simulation and run directly on the physical hand, with no fine-tuning and no state estimation.
- The authors release the mechanical design, simulation model, identified mapping, training environments, and deployment stack.
- Paper: https://arxiv.org/abs/2608.28578