Researchers led by Yiting Chen have trained a humanoid robot to open round doorknobs, a task that needs a firm multi-finger grip, enough twist without slipping, a controlled push and a stable body, according to a paper on arXiv. Their method, Tactile-Driven Motor Adaptation (TMA), builds the learning around contact the robot can feel through touch and adjusts its motor behaviour as it goes, based on its recent tactile history.
A policy for the upper body is trained together with a general walking controller, so that the hands and the whole body work together. Trained entirely in simulation, the system ran without further training on a Unitree G1 humanoid with BrainCo Revo2 hands and clearly beat standard Rapid Motor Adaptation and reactive tactile baselines on four commercial round doorknobs.
Background
Unitree G1
Humanoid robot
The Unitree G1 is a small humanoid robot from China's Unitree Robotics, and research teams, including one at Apple, use it to test new ways of training and controlling humanoids.



