Tactile robot finger outperforms humans in identifying textures


June 18, 2012

The BioTac sensor can correctly identify a randomly selected material from a a sample of 117 common materials 95 percent of the time

The BioTac sensor can correctly identify a randomly selected material from a a sample of 117 common materials 95 percent of the time

Image Gallery (4 images)

We’ve seen the development of a number of technologies that could be used to provide robots with a sense of touch, such as proximity and temperature sensing hexagonal plates and artificial skin constructed from semiconductor nanowires. However, perhaps none are as impressive as a tactile sensor developed by researchers at the University of California’s Viterbi School of Engineering. The group’s BioTac sensor was built to mimic a human fingertip and can outperform humans in identifying a wide range of materials, offering potential use for the technology in robotics and prostheses.

The finger-sized BioTac sensor consists of a soft, flexible skin covering a liquid filling. The skin even has fingerprints that enhance the sensor’s sensitivity to vibration, so that as the sensor slides over a textured surface, it vibrates in characteristic ways. A microphone designed to be used underwater – known as a hydrophone – located within the sensor’s bone-like core detects these vibrations, similar to the way in which a human finger uses similar vibrations to recognize textures. However, the robot finger is even more sensitive and also has the capability to identify in which direction forces are being applied and the temperature of the object being touched.

The researchers also developed a new algorithm that mimics the exploratory movements humans make based on prior experience with similar objects when attempting to identify an object by touch. Previously, there had been no way to decide which exploratory movement to make next, but Professor of Biomedical Engineering Gerald Loeb and recently graduated doctoral student Jeremy Fishel came up with a new theorem for solving the general problem which they describe as “Bayesian Exploration.”

The specialized robot, which was built by Fishel, was trained using 117 common materials sourced from fabric, stationary and hardware stores. After making an average of five exploratory movements, the robot could correctly identify a randomly selected material with 95 percent accuracy. It was only occasionally stumped by pairs of similar textures that human subjects, making their own exploratory movements, couldn’t distinguish at all. Fishel and Loeb say that, although their robot finger is very good at identifying which textures are similar to each other, it can’t tell which textures humans will prefer.

Fishel and Loeb are partners in SynTouch LLC, a technology start-up founded in 2008 by researchers from USC’s Medical Device Development Facility that is now selling the BioTac sensor to other researchers and manufacturers of industrial robots and prosthetic hands.

The researchers’ study, “Bayesian exploration for intelligent identification of textures,” is published in Frontiers in Neurobotics. Fishel and Loeb explain their technology in the following video.

Source: University of Southern California

About the Author
Darren Quick Darren's love of technology started in primary school with a Nintendo Game & Watch Donkey Kong (still functioning) and a Commodore VIC 20 computer (not still functioning). In high school he upgraded to a 286 PC, and he's been following Moore's law ever since. This love of technology continued through a number of university courses and crappy jobs until 2008, when his interests found a home at Gizmag. All articles by Darren Quick
Post a Comment

Login with your Gizmag account:

Related Articles
Looking for something? Search our articles