EgoThumb: a sixth robotic finger that guesses through vision when to grasp and hold an object

This extra thumb, driven by a camera attached to the palm and driven by imitation, allows you to hold and manipulate with one hand. Hardware and software are published in open source.

Supernumerary robotic fingers aim to make a hand capable of gestures that normally require two, like holding an object while manipulating it. Existing models often remain cumbersome, reserved for large objects, and trigger on an explicit movement of the hand regardless of what surrounds it.

EgoThumb, presented by a team around Hao Su, bypasses these two boundaries. The finger, activated by two tendons, lifts its motors on the forearm, which lightens the end while keeping accurate joint control. A palm-oriented camera filmes the intended hand and object, and a policy learned by imitation translates these images directly into joint positions: the user's intention is deduced from the visual context, without a dedicated command.

According to the authors, the system is widespread to various users and environments. Material plans and code are fully open.