Google created a program for self-learning robots walking

Widespread use of robots in variousindustries, services, agriculture and transport is acquiring an avalanche-like character. Robotic systems are used almost everywhere. However, for the full integration of mechanical assistants in human life, robots will have to undergo a serious training course, including the acquisition of the most basic skills - independent movement and orientation on the ground. Currently, training on the movement of robots is carried out according to a laborious scheme that requires constant operator control, however, the latest Google developments will significantly simplify the process.

Google Robotics engineers have developed a robot withfour limbs, which in a few hours mastered the basic techniques of moving around the area back and forth, with the implementation of turns. The new algorithm used primarily the real situation, abandoning the traditional creation of a virtual model of the environment, traditional for modern robotics.

Google developers proceeded from the fact thatcreating a computer model of some surfaces, such as gravel, is quite difficult and inefficient. Therefore, robots were given the opportunity to study in real terrain.

Using a methodology close toreality allowed the robots to quickly adapt and learn to walk on rough terrain. However, during testing, operators had to constantly adjust the program. For greater effectiveness, it was decided to limit the space for training. When reaching the border of the territory, the robot independently understood that it was necessary to perform a U-turn and go in the opposite direction.

As a result of the study, four-legged robotsmastered the methods of movement in different terrain and were able to learn to walk without operator intervention. However, for this algorithm, there are some limitations associated with the use of a motion capture system that allows the robot to determine its location. In real terrain, such conditions cannot be reproduced. In the future, according to this technique, robots of various designs and modifications will be trained in a single environment, which will create a large database for the subsequent development of all robotics.

Source: engadget