A glove to teach robots

New study throws lights on the characteristics of human grasp

We can sense the nature, texture, even properties of materials using touch. This ability of identification comes with our understanding of objects with respect to the amount of force applied in touch. A network of sensory receptors called mechanoreceptors, present in our skin are fueling this ability. Humans use this potential to build his dexterity (skilled in performing tasks with hands).

Subramanian Sundaram from the Computer Science and Artificial Intelligence Laboratory at the Massachusetts Institute of Technology was thinking about bringing this dexterity in robots.
He along with his team, attached a sensor array of 548 sensors on a knitted gloves. Palms and fingers of glove contains a force-sensitive film.

Using this glove, the weight and shape of different materials is identified with respect to the change in pressure absorbed by the sensors. The collected data is transfered to the neural network to learn the patterns.

At present, the neural networks are trained to identify only 26 objects and the network is able to identify objects with 89.4 percent accuracy.

The experiment using cheap glove is increasing the possibilities of improved datasets of various objects and the study could bring drastic changes in understanding our own grasping feature as well as in robots.

Journal Source: https://doi.org/10.1038/s41586-019-1234-z
Note: The feature image is for representation only

Written by M R Raghul