A team at the Hong Kong University of Science and Technology (HKUST) has developed the EmArm robotic arm, which senses touch across its entire surface and adapts its actions accordingly. A soft optical "skin" is built directly into the arm's rigid skeleton. The work was published in the journal Nature Sensors.
The skin covers the upper and lower parts of the arm, while miniature cameras inside analyze the distribution of forces. The spatial resolution is 0.085 mm - roughly that of a human fingertip. The two-point discrimination threshold (the minimum distance at which two simultaneous touches are felt as separate) reaches 18.2 mm, whereas for the skin of a human hand it is usually 30 to 50 mm.
The optical principle made it possible to do without the complex wiring required by electronic skin and made the sensor resistant to electromagnetic, light, temperature, humidity and acoustic interference. In experiments, sensing accuracy was maintained at up to 97.9% under five types of external influence.
The soft skin did not degrade the arm's force characteristics: repeat positioning accuracy remained comparable to that of conventional manipulators. The system withstood pressure of more than 15 kg and a torsional load of 7.35 N·m (newton-meter - a unit of torque).
So that the arm would not merely collect signals but understand them, the team proposed a closed "perception - action" loop. It combines data on joint position (proprioception - the sense of one's own body) with tactile information and in real time identifies the contact location, area, force and number of touch points. These data go directly to the motion control system.
In tests, EmArm recognized five types of touch - stroking, pinching, poking, tickling and hitting - with accuracy of up to 99.8% and responded with emotional expressions: joy, sadness, curiosity, embarrassment or anger. When a ball was placed on the arm offset from its center of gravity, it independently adjusted its position and forces to hold it. And when a human hand appeared in the arm's movement zone, detected by an invisible camera, the arm stopped instantly, determined whether the obstacle was temporary or permanent, and rerouted.
The authors note that the approach can be extended to the legs, torso and other parts of robots. This could be useful in home services, industry and medical care, as well as for the development of "embodied AI" - artificial intelligence systems that learn about the world through a physical body. For now, this is a laboratory prototype, not a finished product.