"From neural networks and sticks": how an unusual robot was taught to walk

Today, the robots of the most bizarre forms are veryhard to surprise. However, a group of scientists from Japan seems to have succeeded. They built strange robots from sticks and ... other scrap materials. Moreover, thanks to a very advanced system of artificial intelligence, built on the basis of neural networks, the robot learned to walk. And strange as it may sound, this approach has a very promising future.

How are robots created

Usually, when designing robots, scientistsyou must have a very clear idea of ​​what the robot should do, how it will move and what actions it will need to perform. In this case, you first need to build a prototype of the future device in order to ensure that all their calculations were correct and to identify design flaws. But the development process does not end there. After forming an understanding of how the model will function, it will be necessary to “teach” it to those actions that it must perform, be it as a manipulator, walking or moving in other ways.

Branch robots

But, as scientists from Tokyo sayUniversity and Preferred Networks, the process of creating robotics doesn't have to be that complicated. Back in December last year, they began experiments on the creation of robots based on servo drives and available materials (for example, tree branches).

See also: Robots have learned to recognize objects with a glance and touch

At the same time, even before creating a model, robots are trainedmovement in a special computer simulation, where an advanced neural network teaches them to walk based on deep machine learning. In order to upload model data to a neural network, scientists simply take several branches of a suitable size and the body of a robot. All this is subject to 3D scanning. The resulting models appear in a special computer program that “collects” them together and begins the learning process.

</ p>

During these “lessons,” a behavior pattern is defined.future robot. The neural network is “rewarded” by the future mechanism for the correct movements, and “punished” for the incorrect ones. In addition, there is the possibility of manual adjustment of movements, as well as a system that provokes the occurrence of random events during training so that the robot is fully prepared for what awaits it in the real world. When training ends, it remains only to assemble the model and load the behavior algorithm into it.

The robots are controlled by a system based on the Arduino Mega and servo drives Kondo KRS-2572HV

Why create such robots?

In fact, the potential of such a system is simplyis huge. In fact, all you need is a “computer and a pair of motors.” The rest of the details can be collected right under your feet. Such robots can be used for reconnaissance in regions where a person cannot reach, and sending full-fledged mechanisms is too expensive. In addition, these robots can be collected, for example, during the colonization of other planets. You can, for example, send an assembly module with a stock of microprocessors and servos, and other details can be assembled upon arrival at the mission site.

Discuss this and other news you can in our chat in Telegram.