This robot dog learned to stand up after being knocked down

AI reinforcement learning could allow robots to handle a wide variety of tasks and situations.

Yang et al

At some point when you were a young child, you learned to get up after a fall and eventually to walk. You probably received encouragement from your parents, but for the most part, you learned through trial and error.

This is not how robots like Boston Dynamics Spot and Atlas learn to walk and dance.

They are meticulously coded to tackle the tasks we set for them. The results can be impressive, but they can also leave you unable to adapt to situations that are not covered by your software. A joint team of researchers from Zhejiang University and the University of Edinburgh states that ‘In a recent article published in the journal Science Robotics, they detailed an AI reinforcement approach they used to enable their dog-shaped robot, Jueying, learn to walk and recover from falls on your own.

Yang et al

The team told Wired that they first trained software that could guide a virtual version of the robot. It consisted of eight AI “experts” who trained to master a specific skill. For example, one learned to walk fluently, while another learned to balance.

Every time the digital robot successfully completed a task, the team rewarded it with a virtual point. If that all sounds familiar, that’s because it’s the same approach that Google recently used to train its groundbreaking MuZero algorithm.

Related Posts

Once they successfully trained the eight experts, they developed an additional network to act as a sort of head trainer that would manage the inputs of the other eight algorithms, prioritizing one or the other. more as needed in a given situation.

Yang et al

You can see it in action in the GIF above. They then ported their software to one of their prototypes and put it to the test. You can see the punishment he was subjected to when he was kicked and pushed to the ground. , got up, and began to walk again. the march, to handle a variety of different tasks they’ve never seen before.

”However, it may be a while before we see Jueying and Spot battling for the best robot dog award. One of the challenges the team faces is reducing the amount of computational power it takes to simulate robot training, which it will need to do to make it more useful for practical applications.

Engadged / TechConflict.Com

Contact Us