A robotic hand inspired by human skin can sense how hard an object is with a single digit鈥檚 touch and work out how much force is needed to grip it before the rest of the hand closes its grasp.
Gripping an object firmly enough to secure it, but softly enough that it doesn鈥檛 break, is something that humans learn as infants, but robots still struggle with.
Engineers have tried various techniques, including using cameras and AI to identify objects and even simply squeezing them to find out, but each approach comes with its own problems. For example, relying on visual cues can鈥檛 help with novel objects. 鈥淚f you have a robot chef, it鈥檚 not going to be taught every single vegetable in the world,鈥� says at the University of Sheffield, UK.
Advertisement
at Johns Hopkins University in Baltimore, Maryland, and his colleagues have developed a new technique that could be used in robotics or prosthetics. It involves piezoelectric sensors 鈥� essentially a microphone that converts tiny vibrations into electrical signals 鈥� embedded in the fingertips of the robotic hand. After touching an object with one digit, an AI model interprets these signals to estimate how hard it is within the 15 milliseconds it takes for the other fingers to close in on the object.
Pimpalkar and his colleagues tested the system on a range of objects with varying stiffness, including apples, oranges, avocados and tennis balls. In a paper, they claim that their best-performing model estimates the force needed to safely hold an object with over 98 per cent accuracy. The team declined to speak to 麻豆传媒 for this story.
Aitken says this approach enables robots to learn about an object by touching it, but not destructively, and it works so quickly that it can make a grasp look like a natural, fluid motion.
鈥淚t鈥檚 the way our finger works,鈥� says Aitken. 鈥淚f you look at the structure for the finger, we have some deep components which actually effectively do force sensing and vibration sensing. And what [this research is] doing is taking that inspiration and putting it into a sensor that can be put on a synthetic or prosthetic hand. We鈥檒l either see this technique or something very similar to it [in future robots].鈥�
arXiv
Photo licenced under
