A group of researchers at the University of California Berkeley have trained a dog to walk on various terrain in under twenty minutes using machine learning. Using a ‘deep reinforcement’ form of artificial intelligence, the robot can quickly acclimate to traversing almost any ground cover it finds itself on. Obviously, there’s never been a better time to build a treehouse.
Using a Q-learning algorithm, the robot doesn’t require any prior knowledge of the terrain; it just immediately starts the learning process from the ground up, being rewarded for successful actions. Wait – what are they giving it, digital dog treats? How do you reward a robotic dog? I’m so confused right now. Granted, I’m confused most of the time, but right now included.
For reference, it took me a solid three years to learn how to walk. Before that, I could only crawl in reverse and get stuck under the sofa. You never heard my parents complaining, though; apparently, it made me incredibly easy to babysit.
[via TechEBlog]
