A wheeled robot set loose on a college campus has figured out how to open all kinds of doors and drawers while rolling around in the real world.
The robot adapted to new challenges on its own – paving the way for machines capable of independently interacting with physical objects. “You want the robots to work autonomously… without relying on humans to keep giving examples at test time for every new kind of scenario that you’re in,” says Deepak Pathak at Carnegie Mellon University (CMU) in Pennsylvania.
Pathak and his colleagues initially trained the robot through imitation learning, providing visual examples of how to open objects such as doors, cabinets, drawers and refrigerators. They then turned it loose on the CMU campus to try opening doors and cabinets it had never encountered before. This required the robot to adapt to each new object using artificial intelligence that rewarded it for figuring things out.
The robot typically spent from 30 minutes to an hour learning how to consistently open each object, says Haoyu Xiong at CMU, who built the robot and scouted out the campus for a wide variety of test locations. The team included 12 training objects for practice and then eight additional objects as a test of the robot’s capabilities.
Although its initial success rate was about 50 per cent on average, the robot sometimes completely failed to open a new object when first starting out. By the end, its success rate rose to about 95 per cent.
In addition to learning on the fly, it had to be able to physically handle heavy doors, says Russell Mendonca at CMU. Achieving both goals cost $25,000, he says, which is much less expensive than other robotic systems with adaptive learning capabilities.
The robotic demonstration outside the lab “marks a concrete step toward more general robotic manipulation systems”, says Yunzhu Li at the University of Illinois Urbana-Champaign. “Opening doors and drawers – a seemingly simple task for humans – is actually surprisingly difficult for robots,” he says.