A wheeled robot released on a college campus has discovered how to roll around the real world and open all kinds of doors and drawers.
Robots have adapted themselves to new challenges, paving the way for machines that can independently interact with physical objects. “We want the robots to be able to operate autonomously, without having to rely on humans to keep giving them examples of all new kinds of scenarios during testing,” he says. Deepak Pathak at Carnegie Mellon University (CMU) in Pennsylvania.
Pathak and his colleagues initially trained the robot through imitation learning, which provided visual examples of how to open objects such as doors, cabinets, drawers, and refrigerators. They then unleashed it around CMU's campus, opening doors and cabinets they had never encountered before. This required the robot to adapt to each new object using artificial intelligence that rewards the robot for understanding things.
The robot typically spent 30 minutes to an hour learning how to open each object consistently. Haoyu Shion CMU Ph.D. built a robot to scout various testing locations on campus. The team included his 12 training objects for practice, and then he included eight additional objects to test the robot's abilities.
Initial success rates averaged about 50%, but the robot sometimes completely failed to open new objects when first started. Eventually, that success rate increased to about 95%.
In addition to learning on the fly, he said he had to be able to physically handle heavy doors. Russell Mendonca At C.M.U. Achieving both goals will cost him $25,000, which he says is much cheaper than other robotic systems with adaptive learning capabilities.
Demonstrating the robot outside the lab “represents a concrete step towards more general robotic manipulation systems,” he said. Yunju Lee At the University of Illinois at Urbana-Champaign. “Opening a door or a drawer is a seemingly simple task for humans, but it's actually surprisingly difficult for robots,” he says.
topic:
Source: www.newscientist.com
