WPI Research is the research magazine of Worcester Polytechnic Institute. It contains news and features about graduate research in the arts and sciences, business, and engineering, along with notes about new grants, books, and faculty achievements.
Issue link: http://wpiresearch.epubxp.com/i/90696
VIDEO EXTRA Professor Chernova on how robots learn from example. " Being more interactive with the world gives you more information and better sensory feedback enables you to interact better. Robots that are detached from the world can't perform tasks as well." — Eduardo Torres-Jara the modeling has been based on vision systems — which has limitations. Rather than adding more cameras, he might begin by investigating how newborns function before their sight is fully developed — or how an adult would search for the TV remote control in the dark. He began his quest by rethinking robot hands. A typi- cal robot has a rigid claw at the end of its arm that opens and snaps closed, regardless of what it is gripping. Our own fingers, by contrast, are soft, flexible, and laden with nerve endings. Torres-Jara has revolutionized the design and con- trol of robot appendages by making them more like human hands, with soft pads that relay feedback from a multitude of sensors. His novel tactile sensors go hand-in-hand with a more flexible approach to artificial intelligence (AI), one that builds models based on knowledge of the real world, rather that writing algorithms that execute pre-planned tasks. "When we increase sensory data, the software can be written differently, because it's based on the physical environment, with less reliance on pre-planning," he says. Where conventional grippers can crush or deform ob- jects, Torres-Jara's "smart sensors" are sheathed in a flexible skin of silicone rubber, giving them a gentle grip that yields to an object's contours. With a network of sensors that detects forces in multiple axes, the pads enable a robot to sense and manipulate varied objects without having to know their precise location or configuration. Applied to the feet of the robots in his lab, they offer multiple points of contact, improving the robots' ability to navigate uneven terrain. His sensors, which can be mass produced inexpensively and easily from existing electronic components, hold great promise for industrial, medical, and military applications. They can be made firmer, to stand up to factory use, or so [6] exquisitely sensitive they can detect a breath of air, or even "read" the raised lettering on a penny. Their commercial uses could range from auto safety devices to bomb disposal apparatus, Torres-Jara says. When a robot can sense its environment, it doesn't need to be reprogrammed each time the task parameters change. With the "sensitive manipulation" hardware and software that Torres-Jara is developing, every step in an assembly task can be based on tactile feedback, making complex tasks pos- sible — and more reliable. "Now the robot actually knows what is happening," he says, "and if an object moves, the robot can sense that and follow it. Before, with automatic mapping, you had to hope the object was there. "Being more interactive with the world gives you more information," he says, "and better sensory feedback enables you to interact better. Robots that are detached from the world can't perform tasks as well. By enriching the input from sensory feedback, we will move AI forward." Robots for the Rest of Us To succeed in the real world, robots will need more than the ability to sense the environment. They will also need to learn and adapt to changing tasks and user needs, much as living creatures do. Creating this new type of robot is the goal of Sonia Chernova, assistant professor of computer science and robotics engineering, whose research centers on robots that can learn on the job, from their human users — even when the users are not trained experts. Robots that can learn "on the fly" and function in natu- ral, unconstrained environments would be an asset in the home, where they could assist elderly and disabled people, and in factories, to reduce the need for reprogramming when gearing up for new products. "My fundamental belief wpi.edu/+research