DeepMind, a Google AI company, is working with a new robotic hand designed for extreme durability. This hand, developed by the Shadow Robot Company, is built to withstand significant damage while still offering impressive speed and flexibility.

The key to the hand’s toughness lies in its modular design. Each finger is built to be easily swapped out if broken, and the entire hand can be modified to include additional fingers for specialized tasks. This focus on modularity makes the hand more repairable and adaptable than traditional robotic limbs.

Despite its robust build, the hand boasts impressive dexterity. It can move from fully open to closed in under half a second and exert a forceful pinch. This combination of speed, strength, and durability makes the hand ideal for reinforcement learning experiments.

Reinforcement learning relies on trial-and-error, and robots inevitably make mistakes during this process. A standard robotic hand might break easily under such duress, hindering the learning process. This new, nearly indestructible hand allows robots to experiment freely without fear of damage, accelerating the pace of AI development.

The hand also boasts a sophisticated sensor suite. Hundreds of sensors on each fingertip and dozens more along the segments provide detailed information about the objects the hand interacts with. Tiny cameras capture the interior of the hand’s silicone skin, allowing the AI to gauge an object’s hardness and shape based on how it deforms the skin.

This combination of durability, dexterity, and advanced sensory feedback makes the new robotic hand a valuable tool for DeepMind’s AI research. It paves the way for faster development of robots capable of complex manipulation tasks in challenging environments.