In order for a robotic mechanism to learn to walk, it is not enough just to "attach" several legs to it. Learning to move is a very complex process that takes developers a lot of time. But now this issue will be solved by artificial intelligence, because a group of experts have created universal algorithms that help AI teach robots of any configuration to move. In this case, human intervention in this process is not required.
Behind the development is a team of scientists from the University of California at Berkeley and a group of experts from Google Brain, one of Google's research arm on artificial intelligence. Their new system trained the four-legged robot to traverse both familiar terrain and unfamiliar ones.
Reinforcement learning is essentially a carrot and stick method adapted for AI. He uses reward or punishment for achieving or not achieving goals.
For experiments, scientists took the Minitaur robot. They developed a system consisting of a workstation that updated the neural network data, loaded the information into Minitaur, and downloaded it back. The NVIDIA Jetson TX2 chip on board the robot was responsible for information processing. The robot walked for 2 hours and took 160,000 steps. During this time, the algorithm rewarded the robot for moving forward and punished it if it got stuck in place or gave a very large roll to the side. As a result, a movement algorithm was created that allowed the robot to choose the optimal trajectory of movement in any situation.
Vladimir Kuznetsov