Teaching robots how to recognize and handle unknown items remains a key hurdle to enabling automation of a much wider range of routine tasks.
But researchers at the Massachusetts Institute of Technology may have come up with a relatively simple solution: they outfitted an ordinary glove with hundreds of tiny, inexpensive sensors, then used the way its wearers handled things to build a data set of how future robotic arms should respond to a variety of objects.
The scalable tactile glove, or STAG, is a knitted glove laminated with a conductive polymer and threaded with a conductive thread, which resulted in 550 different sensors that captured the pressure applied by a human grip as it interacted with an object.
A neural network then processed the signals, allowing it to identify and analyze 26 common objects, from a soda can and scissors to a pen and a mug.
Using the data set in combination with other systems, such as image recognition or computer vision, could ultimately give robotic arms a more intuitive understanding of unfamiliar objects.
MIT researchers noted that the glove is made with materials costing about $10, and that it also demonstrated how movements in one part of the hand corresponded to movements elsewhere — a potential breakthrough for prosthetics design.