“From neural networks and sticks”: how unusual is the robot taught to walk

Today, robots of different shapes is very difficult to surprise. But a group of scientists from Japan seem to have succeeded. They built strange robots from sticks and… other scrap materials. Moreover, thanks to the very advanced system of artificial intelligence based on neural networks, the robot learned to walk. And as strange as it may sound, but this approach has a very promising future.


How are robots

Usually, when designing robots, scientists need to have a very clear idea of what the robot should do, how he moves and what actions it should perform. In this first required to build a prototype of the future device in order to ensure that all their calculations were correct and to identify deficiencies in the design. But the development process does not end there. After the formation of understanding about how to operate the model, you will need to”teach” it the actions that it should perform, whether you are working as a manipulator, walking or moving in other ways.

Robots from the branches

But as stated by scientists from the University of Tokyo and the company Preferred Networks, the process of creating a robotics does not have to be so difficult. In December of last year, they began experiments to create robots based on servo drives and scrap materials (e.g., tree branches).

See also: Robots have learned to recognize objects using sight and touch

However, before the creation of the model, the robots learn to move in a special computer simulation, where they learn to walk advanced neural network-based deep machine learning. In order to download the model in the neural network, the scientists just take a few branches of the right size and the robot body. All this is subjected to 3D scanning. The models appear in a special computer program, which collects them together and begins the process of learning.


During these “lessons” sets the model of the future behavior of the robot. For a true movement neural network “rewards” the future mechanism, and for incorrect — “punishing”. In addition, there is a manual setting of movements, as well as the system, which provokes the occurrence of random events during training in order for the robot was the most prepared for what awaits him in the real world. When training ends, it remains only to assemble the model and load the behavior.

The robots are controlled by the control system based on Arduino Mega and servos are Kondo KRS-2572HV

Why create such robots?

In fact, the potential of such a system is huge. In fact, all you need is a “computer and a couple of motors”. Other details can be collected directly at his feet. Such robots can be used for exploration in the regions where man can not reach, and send the full mechanisms are too expensive. In addition, such robots can be assembled, for example, when colonization of other planets. You can, for example, to send the Assembly module with the supply of microprocessors and servos, and other items to collect on arrival at the place of the mission.