New Google algorithms will allow robots to learn to walk on their own

Google has conducted a study and set up algorithms that will speed up and simplify the process.


What did

Researchers from Google Robotics have createdfour-legged robot, which for several hours he himself learned to walk back and forth and turn. Previously, the algorithms were tested on a virtual robot in a virtual environment, which helped prevent damage to a real robot. But the process of emulating different surfaces was too time consuming.

</ img>

Researchers began training in the real world andthe robot was able to adapt more quickly to uneven terrain. However, human intervention was still necessary, and researchers had to intervene in his training. Therefore, they limited the surface area and forced the robot to learn several maneuvers simultaneously. If the robot, reaching the edge of its territory, goes forward, it will understand that there is an obstacle ahead and will begin to go back, acquiring a new skill.

Using this system, the robot was able to use the trial and error method to ultimately learn how to independently navigate on different surfaces.

However, the study has limitations: until the technology is universal and suitable for all types of robots. But researchers hope to adapt new algorithms to different robots. If they succeed, it can greatly advance robotics.