The robot created by scientists taught to play Jenga
Artificial intelligence is becoming more aware. In particular, the robot created by scientists can independently play Jenga. This is a huge progress for technology, because it shows that machines have learned to touch and control their actions. Thus, the researchers hope to teach the development of object management in the real world. This became known from an article published by experts at the Massachusetts University of Technology in Science Robotics.
The robot created by scientists will not beat you in Jenga yet
For training the robot, the professionals developed a special sensor with a two-contact manipulator. After that, the car was offered to play Jenga. The goal of scientists was not victory, artificial intelligence had to study the game model and extract blocks at random.
Nima Fazeli, head of research, noted that the robot is aware of what kind of blocks they have and how they are located. However, initially the machine could not understand how interaction is provided. After some time, the AI revealed certain patterns in the arrangement, including that some blocks are looser than others. It is not necessary to apply too much power to extract them.
Like people playing Jenga, the robot also relied on touch. To the human eye, the tower also looks monolithic, and it seems that any element can cause its collapse. However, it is worth trying these or other parts to understand which “brick” can be removed safely.
Artificial intelligence relied on trial and error, and thus mastered interaction with it. Fazeli stressed that awareness came about 200 elements. The researcher noted the machine was trained like a small child. After all, each of us got acquainted with the basic concepts and mechanisms of physics not at school, but at home, thanks to the games.
Scientists at the University of Massachusetts were the first to try out this method of teaching the robot to touch. Researchers at other institutions have tried alternatives. In particular, the University of California at Berkeley decided to reward the car when it performs the task. However, in this way only a solution can be achieved, but not an understanding of how the AI did it.
Of course, Jenga is not the most important skill for combining vision and touch. But for the robot it is extremely important. Researchers compared this with daily teeth cleaning. That is, a person visually understands what he is doing. But at the same time for cleansing it is necessary to apply a certain force. Of course, it is impossible to determine it with the help of sight. The same principle is relevant for cars. If automated teeth cleaning systems are ever developed, artificial intelligence must not only see what it does, but also feel the necessary strength.
The method applied by researchers at the University of Massachusetts, clearly demonstrates how the ways of teaching machines have changed. Previously, researchers could only simulate situations. And this method showed a certain efficiency. However, to simulate tasks and perform them in the real world is not the same thing. Jenga allowed the robot to realize the physics for what it is on our planet, and not in a computer simulator.