At some point, when you were a child, you learned to get up after falling and eventually to walk on your own feet. You probably received encouragement from your parents, but most of the time, you learned by trial and error. This is not how robots like Spot and Atlas from Boston Dynamics learn to walk and dance. They are meticulously coded to handle the tasks we are assigned to them. The results can be impressive, but they can also make them unable to adapt to situations that are not covered by your software. A joint team of researchers from Zhejiang University and Edinburgh University claims to have developed a better way.
Yang et al
In a recent article published in the magazine Robotics Science, they detailed an AI reinforcement approach they used to allow their dog-like robot Jueying to learn to walk and recover from falls on its own. The team said Wired they first trained software that could guide a virtual version of the robot. It consisted of eight AI “experts” that they trained to master a specific skill. For example, one became fluent in walking, while another learned to balance. Each time the digital robot successfully completed a task, the team rewarded it with a virtual point. If all of this sounds familiar, it’s because it’s the same approach that Google recently used to train its innovative MuZero algorithm