Bacteria do it. Viruses do it. Worms, mammals, even bees - everyone does it. Every living thing on Earth is reproduced, whether by the asexual (boring) or sexual (fun) method. Robots do not. Machines made of steel are not very interested in reproduction. But maybe they can learn. Scientists from the field of evolutionary robotics are trying to make machines adapt to the world and, ultimately, reproduce on their own, like biological organisms.
For example, someday two robots thatespecially well adapted to a particular environment, they can combine their genes (ok, code) to create with the help of a 3D printer a small child robot that will have the power of its two ancestors. If this approach works, it can lead to the appearance of robots that construct themselves, creating perfectly adapted morphologies and behaviors that human engineers never dreamed of.
It seems strange and a little disturbing, butevolutionary robotics are already engaged in such fantastic projects. Australian engineers, for example, developed robotic legs last year, first randomly generating 20 forms. During the virtual simulation, they checked how well each of them would walk on different surfaces, that is, they checked the “suitability” in terms of survival of the fittest. Then they took the best performers and “mated” them to produce similar legs — that is, children. Researchers did it again and again, generation after generation, and created legs that were wonderfully adapted for walking on hard ground, on gravel or on water. The projects are crazy - like wood people, dancing dances from Fortnite (good for solid ground), and oddly deformed elephant legs (good for water).
What is the main idea? Traditionally, when engineers start designing robots, they tend to use old ideas. Why do rovers have six wheels? Because the six-wheeled machines have worked well on Mars before. However, perhaps the designers have missed something. The beauty of evolution is that it constantly runs into insane ideas. No one, for example, has developed a fungus for penetrating and controlling ants in the rain forest - this unusual strategy emerged thanks to the generation of random mutations and natural selection.
As in nature, it is the mutations that will determinethe evolution of robots. Variability is important. When two bodies make a child, their genes unite, but mutations also penetrate into them, which can lead to the appearance of unique features in the child, such as a slightly modified pattern on the wings. Such mutations make the offspring more or less adapted to the specific environment. If this is an unfavorable mutation, the animal does not reproduce as effectively (or does not reproduce at all), and these mutant genes are not passed on to the next generation.
See what the scientist is doing in the fieldComputers Gush Aiben from Amsterdam Free University. He takes two relatively simple robots, consisting of connected modules, and combines them, combining their "genomes", which carry information, say, about coloring. It also adds noise to this combination of data that mimics a biological mutation, slightly changing the offspring so that it is not just a mixture of parents. “One parent is completely green, the other is completely blue,” says Eiben. “A child will have some modules blue, some green, but the head is white. This is a mutational effect. ”
And with this change, a new species appears.creativity in robotic design. “It gives you variety and the opportunity to explore areas of the design space that you don’t usually enter,” says researcher David Howard, who developed an evolving system of legs and recently published a work on evolutionary robotics in Nature Machine Intelligence. "One of the things that makes natural evolution powerful is the idea that it can really adapt a creature to its environment."
The idea is for robots to adapt toniches in a specific environment in a similar way. Suppose you need a robot that can independently explore the jungle. This means that he needs algorithms that control how he moves through vegetation, as well as a morphology that is suitable for thick forests (therefore, no rotors). First you have to model this environment for navigation, select and select those robots that do the job best, and then design slightly modified physical machines based on this.
"As a result, we got a lot of small robots,which are simple and cheap to make, ”says Howard. "We will send them and some will be better than others." If the robot does not return, then it "does not fit" - natural selection in action. Those who cope, will start a new generation, which will automatically be released on a 3D printer. Thus, robotic species evolve. Howard believes that such systems will be spread in 20 years.
By the way, about 3D printers
The materials from which these robots will be made,pose a small problem. “If 3D printing develops faster, this idea will become a reality, but modern printers work very slowly,” says Juan Cristobal Zagall, who studies evolutionary robotics at the University of Chile. Both machines and printed materials are very expensive. But 3D printers can already work with a variety of materials, including metal, and this will make them faster and cheaper.
In general, the range and scope of theserobots will largely depend on how these evolutionary systems display creativity in working with materials. By making ordinary robots, engineers know which materials to use, from metal in engines to carbon fiber in limbs — and this knowledge has developed over many decades of research. However, evolutionary robots open up a new approach to the use of materials. Perhaps a plastic foot would be better when walking in a certain environment than carbon fiber. If the robot survives, then there is something in the combination of components and materials that made it suitable for work, or, in an evolutionary sense, in its niche.
“Perhaps if a new class of materials appears, we can simply plug it in and it can choose,” says Howard. "A human designer will have a lot of time released."
Do you think evolutionary robots have a future? Let's discuss in our chat in Telegram.