A group of researchers from the Massachusetts Institute of technology presented a learning system for robots called C-LEARN. The new development allows robots to not only acquire new skills but also to transfer experience to other robots, making the learning process faster and more efficient. In addition, thanks to C-LEARN “set vector” learning can even people do not know anything about programming.
For the development experts of the laboratory of Informatics and artificial intelligence (Computer Science and Artificial Intelligence Laboratory, CSAIL abbreviated). To begin the process of training a person needs to “show” the robot how to perform a particular job, and ask basic rules to be followed by the robot. Everything else the robot will do, and then will be ready to transfer his experience.
System C-LEARN is quite simple. It is to provide information on end goals and restrictions that will apply in the course of its achievements. For example, if the robot needs to install in the car steering wheel, the operator must specify not only the sequence of actions (what and where to tie), but also to configure several parameters according to which the robot will operate. That is, he will have to distinguish the intended subject (since they are both round in shape), fastening elements and so on. Then the robot performs a task, and people either approve of the result and puts it into a database, or making the necessary adjustments. In addition, the system adopts two kinds of training: the above learning through examples, and learning algorithms using high-level programming.
After the robot acquires the necessary skills, they can be transferred to another machine. During a series of experiments it was found that in the case of a robot man, the learning process was successful in 100% of cases, but when the robot was taught by another robot, the success rate was only at 87.5%. However, during the tests, the staff trained CSAIL robot Optimus to disarm a certain type of explosive device, after which Optimus gave this experience another robot called Atlas and both robots have used this skill with the same efficiency.
Presents a system that allows the robot to teach other robots
Vladimir Kuznetsov