If the robot grabs delicate objects, the bot needs to know what those objects are so it can handle them accordingly. A new robotic hand makes this possible by sensing the shape of objects along the length of three fingers.
Developed by a team of scientists at MIT, this experimental device is known as the GelSight EndoFlex. As the name suggests, it incorporates his GelSight technology from the university, which was previously only available in the finger pads of robotic hands.
The EndoFlex’ three mechanical fingers are arranged in a Y shape. It has two “fingers” on top and an opposable “thumb” on the bottom. Each consists of an articulated rigid polymer skeleton encased in a soft, flexible outer layer. The GelSight sensors themselves (two per digit) are located on the top and underside of the central section of those digits.
Each sensor incorporates a slab of clear synthetic rubber coated on one side with a layer of metallic paint. This paint will act as the skin for your fingers. When paint is pressed against a surface, it deforms to the shape of the surface. Viewed from the unpainted side of the rubber, a small built-in camera (with the help of three-color LEDs) images fine contours of the surface and can be pressed into the paint.
Special algorithms in linked computers transform these contours into 3D images, capturing details less than 1 micrometer deep and about 2 micrometers wide. Paints are needed to standardize the optical quality of surfaces so that the system is not confused by multiple colors and materials.
For the EndoFlex, images from six sensors at once (two for each of the three fingers) can be combined to create a three-dimensional model of the item in your grasp. Machine-learning-based software will be able to identify the object represented by the model after the hand has grabbed the object just once. The system has an accuracy of about 85% in its current form, but this number should improve as the technology develops further.
“Having both soft and hard elements is very important for any hand, but it is also important to be able to perform good sensing over a very large area, especially if you have a very complex hand that you can do with your own hand. Especially if you want to consider performing manipulation tasks.” Sandra Liu, a mechanical engineering graduate student who co-led the research with undergraduate Leonardo Zamora Yáñez and Professor Edward Adelson.
“Our goal with this work was to combine everything that makes the human hand so great into a robotic finger that can perform tasks that other robotic fingers cannot currently do. “
Source: MIT
