Scientists at ETH Zurich have created a functional robotic hand capable of walking on its fingertips, drawing comparisons to the independent appendage "Thing" from The Addams Family. This development allows the robotic hand to move autonomously across different terrains and perform tasks such as pushing objects and pressing keyboard buttons.
The researchers started with a commercially available WUJI robotic hand and added a battery, a small computer, and a motion sensor. This integration created a self-contained unit that carries its own power and computing capabilities.
Teaching the hand to walk involved overcoming challenges posed by the fingers' varying lengths and range of motion. The team utilized reinforcement learning, allowing the control system to practice and refine movements through extensive computer simulations. This digital training was then combined with real-world measurements of the hand's joints and finger slippage to ensure successful physical execution.
The resulting robotic hand can traverse 14 different surfaces, including grass, gravel, hardwood floors, and stone. It also demonstrated the ability to right itself after a fall, succeeding in 21 out of 25 attempts. With the aid of an overhead camera for spatial information, the hand could push blocks to within 17mm of a target destination.
One of the most notable capabilities is the hand's ability to operate a keyboard remotely, pressing individual keys with one finger while using the other four for balance. Researchers even used the hand to play a simple video game by inputting commands onto a separate keyboard, achieving a 29 out of 32 correct key presses.
Lead author Amirhossein Kazemipour envisions future applications where such independently mobile robotic limbs could access confined spaces that larger robots cannot penetrate. A detached hand could crawl into narrow openings to inspect, operate controls, or move objects before being retrieved.
While currently remotely controlled, the technology could pave the way for more complex autonomous robotic systems where specialized parts can function both independently and collaboratively.