Because of their strength and speed, robotic arms are typically permanently attached to floors or other structures for added stability, limiting their reach. The aim of the research, conducted at EPFL's Learning Algorithms and Systems Laboratory (LASA), was to develop a bimodal robotic hand with enhanced grasping capabilities, including occasional independence from the robotic arm to which it is attached.
Robotic hands are usually designed with one goal in mind: to hold things. To create one that can do this and crawl around on its own, like the Addams Family's Thing, researchers created and refined a basic design using a genetic algorithm (which relies on biological tricks like natural selection and evolution) and the MuJoCo physics simulator for testing the practicality of iterations.
The algorithm and simulations helped researchers determine the optimal position and number of movable fingers needed, which turned out to be five, in an arrangement similar to human hands. The robotic hand also has a magnetic connection at the wrist, allowing it to autonomously attach and detach from an arm.
The fingers of the hand can be bent in either direction, allowing some of them to be used to lift objects while the rest act as tiny legs. This design also extends the utility of the hand when attached to a robotic arm. It can lift multiple objects at once without having to rotate the arm to reposition unused fingers.
The hand is also significantly smaller than robots like Boston Dynamics' Spot, which can move freely on four legs. Spot has already been upgraded with its own robotic arm and gripper, but with a movable hand that works independently, it could be better equipped to explore or analyze areas that Spot can't squeeze into.