Scientists at Cornell University’s Organic Robotics Lab have developed a robotic hand that has a level of sensitivity that approaches a human’s — it’s sensitive enough to detect the shape, softness and overall texture of what it touches
The silicone hand — developed by a team of roboticists led by Robert Shepherd, an assistant professor of mechanical engineering at Cornell — is filled with optical fibers that can detect how light that passes through the hand changes as it moves and comes into contact with other objects.
Watch the soft robot pick the ripest out of three tomatoes, in various stages of maturity, just by touching them.
Unlike most tactile robots that feel with sensors on the outside of the machine, this robotic hand has its sensors on the inside, more similar to a human. Typically, robots that sense what they touch rely on the object being able to conduct electricity, which the robot could then detect to try to learn things about what it’s touching.
The hand made by the Cornell lab only works when the machine moves and changes shape. “If no light was lost when we bend the prosthesis, we wouldn’t get any information about the state of the sensor,” Shepherd said in an interview with the Cornell Chronicle. “The amount of loss is dependent on how it’s bent.”
The Cornell researchers say this technology may one day power prosthetic hands that restore people’s sense of touch, or could give biologically inspired robots a more delicate and sensitive style of physical contact.
Huichan Zhao, the Cornell doctoral candidate who is the lead author on the research on the soft robo-hand published in the Science Robotics journal this month, told NPR that she estimates that her team’s soft robotic hand could be made for as cheap as $50.
But, as with other soft-robot projects, the hand from the Cornell lab needs to be filled with compressed air to cause the fingers to balloon, bend and hold their shape. And right now, air pumps are generally too big for a person to comfortably wear.
This article originally appeared on Recode.net.