Robotic hand from MIT can recognize objects by feel

Engineers at MIT have built a three-fingered robotic hand that can identify and safely grasp delicate objects by relying on an increasingly popular approach to making robots useful: making them soft.

Human hands are not easy for robotics engineers to emulate. The simple act of picking up an item involves all kinds of abilities that humans don't notice. Among other things, our grip has to be secure without crushing the thing we're grasping, and our fingers have to form shapes that can fit many types of objects — everything from a sheet of paper or a piece of fruit to a pencil or a living thing.

Researchers at MIT's Computer Science and Artificial Intelligence Laboratory designed a soft silicone "hand" with embedded sensors that they can train to recognize different things. The team will present its research at this month's International Conference on Intelligent Robots and Systems in Hamburg, Germany.

How to play the slide in big technology stocks
Technology IPOs at lowest since 2009

Other researchers and engineers have experimented with soft robotic parts, and this study's co-author, Daniela Rus, has done previous work with soft robots.

A company called Soft Robotics designs soft robotic grippers for commercial purposes, such as packing vegetables into boxes. There's even a soft-robotics toolkit for builders. The MIT design shares similiarities with another three-fingered robot hand built by Carnegie Mellon researchers, which was announced at the IROS 2015 conference this week.

What makes the MIT research unique is the integration of sensors in the hand that can measure the shape of the object, using an algorithm that allows the hand to distinguish one object from others. As a result, the hand can pick up and then identify a wide variety of different objects: a CD, an egg or a cup, for instance.

"The idea with a soft hand is that it's a lot easier for the robot to pick something up," said Bianca Homberg, a graduate student in MIT's Distributed Robotics Lab, and the lead researcher on the project.

"With a rigid hand," Homberg told CNBC, "there has to be a lot of complicated grasp planning to figure out how exactly it is going to pick up the object — where it is going to put its fingers so that it doesn't drop [the object]. With a soft hand you just grab it and the fingers bend around the object and pick it up."

Meet Google's 'Atlas' robot

A "soft hand" allows the robot to perform a greater range of tasks using the same basic strategy. It also behaves in a way that seems more natural — it works in a way that a human hand might. It, for example, can pull a CD off a table in a manner eerily familiar to anyone who lived through the age of compact discs (seen in the attached video.)

The hand has channels in it that fill with air and allow the hand to wrap itself around an object. On-board sensors enable the robot to identify the object it's grasping.

There are limitations, of course. Engineers have to train the hand to recognize each object it's picking up. They make the robot pick up a new object 10 times and then encode that training information in the robot's software. The robot is not yet able to learn how to distinguish new objects by itself and learn as it goes.

"Making robots able to pick things up more easily is one of the things that is very important in pulling them out of the factories and bringing them into the places where they will interact with humans more," Homberg said. "If robots are in human environments, they are going to need to deal with human objects more."