The virtual creature waves its four tentacle-like arms as it propels itself forward. It's creeping up
Graduate student studying computer vision atStanford University in California, Agrim Gupta is not exactly the creator of these strange creatures. He created the ancestors that gave rise to these creatures - unimals, which means "universal animals." This term reflects the fact that they can evolve into very different forms. Some of them resemble real animals. Others are quite strange.
The team found that the body typeanimal affects its ability to learn new things. Scientists tend to believe that even though learning takes place in the brain, it is the body that influences exactly how this process takes place.
If robots could evolve in simulation, they could develop their own shapes that would work even better than those created by humans.
Thus, generalist animals emerged that learned to move in more complex simulated worlds, and eventually developed bodies more suitable for learning.
“I was excited about this job,” says Sam Kriegman. He did not take part in these studies, but he knows a lot on this topic. The scientist works on evolutionary robotics at the Wyss Institute.
Robotic engineers tend to copy the bodies they see in nature. This is why many robots resemble real animals.
To begin with, scientists gave animals bodies that were verysimilar to animal figures. Each has one round head. Straight segments protrude from this head. They branch into other segments, forming body parts that resemble arms, legs, or tentacles. Just over 500 randomly generated animals are thrown into a virtual world much like a video game. In the simplest game, each unital must cross a flat landscape. It figures out how to move using a machine learning computer model.
First, when the model knows nothing about the movement,the body is just spinning, trying random movements. If one movement brings the animal closer to its goal of crossing the landscape, the model learns to repeat that movement. The further an animal travels across the landscape, the higher its score in the game.
Later the animals are divided into groups of fourcopy. The group member who scores the most points gets the opportunity to develop further. Let's imagine that the winner looks a little like a starfish. As he develops, his body changes randomly. For example, he may lose part of his legs. Or all of its legs may grow a new segment. Or one element may become longer and another shorter.
Later, all animals from the original group fromfour people return to the flat virtual world along with a new starfish. They don't remember anything from their first "journey". They all have to start from scratch, spin around until something works. Again, they all get points and compete in groups of four to see who progresses next.
This process is repeated over and over again.Whenever a new animal is created, the oldest one dies. If it did its job well, it would evolve several times before dying. This means that it left behind a bunch of “children and grandchildren” who can become even better. Over many generations, animals get better and better at moving around the landscape. They don't remember anything from past experiences.
The process is then repeated on a more complexterrain. And after and in more difficult conditions - a world with obstacles, external influences and tasks (for example, throwing a ball). The result is specimens that are good at coping with a variety of tasks and moving across any type of terrain.
At the same time, the researchers also laid down a number of rules for “evolution”: symmetrical sides of the body, no more than 10 limbs, which cannot branch more than twice.
The engineers add that translating the simulated creatures into reality will not be easy. The real world is much more confusing and complex than the simulation.
Read more:
Scientists have found a "Pandora's box" in the bowels of the Earth: the energy from there feeds life on the planet
The main myth about dinosaurs has been refuted: scientists have understood how reptiles took over the planet
350 million years ago, something strange happened to the Earth: it affected habitability