A wheeled robot unleashed on a university campus has figured out how to open all sorts of doors and drawers while rolling around in the real world.
The robot independently adapted to new challenges – paving the way for machines that can independently interact with physical objects. “You want the robots to work autonomously… without relying on humans to constantly provide examples of every new scenario you find yourself in during testing,” says Deepak Pathak of Carnegie Mellon University (CMU) in Pennsylvania.
Pathak and his colleagues first trained the robot through imitation learning, providing visual examples of how to open objects such as doors, cabinets, drawers and refrigerators. Then they set it loose on the CMU campus, attempting to open doors and cabinets it had never seen before. This required the robot to use artificial intelligence to adapt to each new object and be rewarded for figuring things out.
The robot typically took between 30 minutes and an hour to learn how to consistently open each object, says CMU's Haoyu Xiong, who built the robot and scoured the campus for a variety of test locations. The team included twelve training objects for practice and eight additional objects for testing the robot's capabilities.
Although the initial success rate averaged about 50 percent, the robot sometimes completely failed to open a new object when first launched. In the end, the success rate rose to around 95 percent.
In addition to learning on the fly, it needed to be able to physically handle heavy doors, says CMU's Russell Mendonca. Achieving both goals cost $25,000, which is much cheaper 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 of the University of Illinois Urbana-Champaign. “Opening doors and drawers – a seemingly simple task for humans – is actually surprisingly difficult for robots,” he says.
Subjects:
