According to the post, the hand acts on its own: it adjusts its movement to disturbances rather than simply running through an uninterrupted sequence of movements.
The post says the system learned by trial and error (reinforcement learningA learning method in which an agent discovers by trial and error which actions bring it the most reward.Read the glossary entry) while following reference movements. This skill was acquired in a computer simulation, then transferred to the real hand.
Source
- Humanoid.guide · humanoid.guide