Robots are not only getting more powerful sensors and processors but also smarter robot arms. Take the Sharpa Wave for instance: it is a robotic hand with 22 active degrees of freedom. It is capable of lifting multiple cans of soda with its fingertip. This video shows a robot using these robotic hands to peel an apple. Researchers used a shared-autonomy system to pull this off:
rather than controlling every finger, the operator triggers pre-learned skills like a “rotate apple or tennis ball” primitive via a keyboard press or pedal. This makes scalable data collection and RL training possible.
They used a Mixture of Dexterous Experts approach to combine vision, language, force, and touch data. In this case, humans provide high level guidance while the robots execute these complex tasks.
[Ht]
