Researchers at Facebook have developed DIGIT – a tactile sensor that will increase the efficiency of the robot arm. She will be able to perform more tasks related to manipulation. Development information appeared in arXiv.
One of the factors limiting modern robotic manipulation systems is the problem of determining contact forces. As a step towards improving robotic manipulations, we present DIGIT – an inexpensive, compact, high-resolution tactile sensor designed to improve the process of manipulating a robotic arm.
Quote from a scientific article
The sensor, developed by a team of researchers at Facebook, works on the basis of sensor technologies developed in the past. The sensor is small and can be mounted on a robotic arm with several fingers.
In order to evaluate the performance of DIGIT, the researchers applied a sensor to each finger on the Allegro Robor Arm – a multi-finger hand developed by Simlab, which is often used to test new methods of manipulating objects. During the experiment, the team trained the neural network model to act as a controller for the Allegro arm: a robot arm moved glass balls.
In these tests, the sensor worked well, it collected high-resolution tactile data that could be used to train deep neural networks in order to better control the Allegro arm.
In the future, researchers plan to continue working on the sensor, they want to make its size even smaller, and also include curved omnidirectional sensitive fields in it.