Google DeepMind explains that useful robots need the hand dexterity people normally take for granted. A multi-fingered robot hand may need to coordinate 22 joints, while simpler grippers still require precise contact and an understanding of an object's orientation in three-dimensional space.
The demonstrations ask a humanoid robot to place grapes in a zip bag and close it, unscrew a spherical light bulb, reorient tools before placing them in a kit, and tie off a trash bag. Each example combines different demands such as twisting, pushing, pulling, pinching and coordinating both arms.
The researchers present dexterity as a prerequisite for handling complex objects outside a laboratory. They highlight potential applications where robots could manipulate hazardous material or complete other intricate work without exposing people to the same risk. Background music and routine demonstration praise are omitted.
Watch on YouTube
