Represented by AI, which alone teaches robots to walk

To robotized mechanism I have learned to walk, not enough just to "attach" to it several feet. mobility training - a very complex process that takes away from the developers a lot of time. But now this issue will be solved artificial intelligence, because the group of experts created a universal algorithms that help robots to teach the AI to move any configuration. In this case, human intervention in the process is not required.

for the development is a team of scientists from the University of California at Berkeley and a group of Google Brain experts, one of the research units of Google for Artificial Intelligence. Their new system has trained four-legged robot to cross as the familiar terrain and unfamiliar.

"Deep reinforcement learning can be used to automate a number of tasks. If we can teach the robot's gait from the ground in the real world, we can create controllers, which are ideally adapted to each robot, and even to individual landscapes, allowing to achieve better maneuverability, energy efficiency and reliability. "- scientists said.

Reinforcement learning - it is, in fact, the carrot and stick method adapted for AI. It uses a reward or punishment in achieving or not achieving the goals.

"Deep reinforcement learning is widely used to train the AI, and even for the transmission of data the real robots, but this necessarily entails some loss of productivity due to inconsistencies in the modeling and requires active intervention. The use of these algorithms in real time proved a daunting task. "

For the experiments, the scientists took Minitaur robot. They have developed a system consisting of a workstation that is updated data of the neural network to download information in Minitaur and discharged back. NVIDIA Jetson TX2 chip on board the robot is responsible for processing information. The robot walked for 2 hours and made 160 000 steps. During this time, the algorithm rewarded for moving the robot forward and punished if it was stuck in place or giving a very large bias towards. As a result, movement of the algorithm was created that allowed the robot in any situation to choose the optimum path.