A team at the University of California at Berkeley used an algorithm called Dreamer to let a robotic dog learn to walk, roll over, and navigate some obstacles in around an hour. The team says the robot wasn’t shown what to do with a computer program but learned everything on its own.
Danijar Hafner, an AI researcher at the school, says that “Reinforcement learning will be a cornerstone tool in the future of robot control.” Future work planned on the robot could involve teaching it how to obey orders and find objects in a room.
Source: Daily Mail