🤖 This robot hand walks... like Thing from The Addams Family

A robotic hand has just learned to walk on its fingers, then use them to perform tasks.

In a laboratory at ETH Zurich, in Switzerland, this five-fingered hand moves around without arms or legs. It walks on its fingers, like Thing in The Addams Family. But behind this surprising scene lies a practical question: how can a robot reach a place too narrow for an entire robot?

ETH Zurich’s “Fingers as Legs” robotic hand

Usually, a mechanical hand is fixed to the end of an arm. The arm brings it to an object, and then the fingers grasp it. Here, the researchers have reversed this arrangement. The fingers also act as legs, carrying the hand to the place where it needs to work.

The prototype weighs 818 g, including its battery, small computer and sensors. So it doesn’t need a cable to move around. The real challenge comes from the shape of its fingers: they aren’t all the same length, and the thumb is on the side. Keeping its balance requires precise coordination.

To teach it to move forward, the team first used a computer simulation. The virtual hand repeated movements and received higher scores when it made progress without wobbling too much. The researchers then adapted the simulation to match the real device’s behavior in the physical world. This allowed the machine to learn which fingers to lift and which to keep on the ground.

ETH Zurich’s “Fingers as Legs” robotic hand walks across a lawn near a golf ball.

ETH Zurich’s “Fingers as Legs” robotic hand walks across a lawn near a golf ball.
Credit: Soft Robotics Lab, ETH Zurich.

The tests covered 14 types of surfaces, from hardwood flooring to gravel and grass. The hand could move forward and change direction using information from its sensors. When it fell, it could also get back up: it succeeded 21 times out of 25 attempts.

The most surprising demonstration takes place at a keyboard. The hand has to free one finger to press a key without losing the support of the others. It successfully entered 29 commands out of 32 in a series of tests. It even pressed the keys in a puzzle game, following instructions from a person.

Another demonstration involved moving a small cube to a precise spot. This time, a camera installed above the scene indicated the object’s position. The hand could then approach and push it while remaining upright.

For now, each activity uses a different learning program. The hand can’t yet decide on its own how to carry out an entire mission, and it doesn’t have an onboard camera to recognize its surroundings. The researchers want to bring these skills together. Eventually, a detachable hand could enter a narrow passage, operate a control, then return to its robot.