Evolution in silicon has shown that machines are capable of self-improvement. Selection and selection generation after generation allows the most adaptable specimens to be fine-tuned and selected. Now robots show scientists how animal behavioral strategies could evolve and improve under natural conditions. Searching for food, fighting competitors, natural cues, how they affect daily life and how they were shaped. In a Swedish laboratory. Laboratory of Intelligent Systems. a group of 10 robots competed for food. What came out of it?
The s-bots chosen for the experiment are 12cm in diameter, 15cm high, and have 2 lithium-ion batteries giving about an hour of autonomous existence. The robots are equipped with a 400 MHz Xscale processor, 64 MB RAM and 32 MB flash memory used for data processing, and 12 PIC microcontrollers for low-level processing. A self-assembled Linux is used as the operating system, and communication with the central station is via WiFi. The sensor capabilities of the robot include infrared sensors (15 around the robot and 4 under the robot), power and speed sensors, humidity and temperature sensors, as well as 8 light sensors, an all-around camera, and 4 microphones.
The robots were programmed to search for a "food source, " which was a lightly glowing ring at one end of the arena. They could only "see" this source at close range with their sensors. At the other end of the arena was a darker ring that was considered "poisoned. The robots got points depending on how much time they spent near the food and / or poison sources, thus demonstrating how successful their artificial lives were.
They could also "communicate" with each other. Each robot could project a blue light, which the others could detect with their cameras, and thereby indicate the position of the food. Thus, light is an information carrier for robots. However, after several generations, robots evolved and taught their descendants not just to communicate with light, but also to deceive their rivals.
Robots were able to evolve because each robot was equipped with an artificial neural network controlled by a binary "genome". The network consists of 11 neurons that are connected to the robots’ sensors and 3 neurons that control their two wheels and blue color. The neurons are connected using 33 synapses, and the signal strength of each synapse was controlled by a single eight-bit gene. Thus, each robot has a 264-bit genome that controls how it will respond to information from the sensors.
Artificial evolution took place in a simulated environment Enki where both robots and their sensors were simulated. Then an evolutionary robot framework was used Teem for the evolution of better cotrollers, which were then transferred to real robots.
In this experiment, each round consisted of 100 groups of 10 robots, each group competing for food in a separate arena. The two hundred robots with the highest scores, the most adapted of the population, moved on to the next round. Their 33 genes were mutated randomly with a 1/100 chance that any one bit would change, and the robots were "paired up" with each other to mix the genomes. The result is a new generation of robots whose behavior has been inherited from the most successful members of the previous generation.
In the initial experiment, the robots emitted light randomly. However, generations later, the robots got better… The light began to carry more and more information and the bots began to orient to the light after only 9 generations.
But, the situation with robots was similar to the behavior of real animals, because it is not always in the interest of the robot (animal) to transmit information about the location of food. The food ring has only 8 places for robots, which means that if everyone gathers near the food source they will have to physically push each other. The effect of fighting intensified when the experimenters allowed the genes responsible for emitting blue light to evolve.
What was the end result? If originally robots emitted light randomly and when they gathered around food the light gave away the location of the food. After evolution, the robots became more stealthy. By the 50th generation they were much less likely to emit light near food than anywhere else in the arena, and the light itself became a less important source of information, making it less appealing to robots.
However, the light never became completely useless. As robots became more cunning and less reliant on light, individuals who glowed near food could be a defense of sorts, because light could be judged as true or false depending on the perception of the other robot. Therefore, the evolutionary pressure to give no signals at all was diluted by the aforementioned factor and because of this, light was still used generations later.
It also means that robots were diverse in their behavior. With free natural selection, processes such as genetic drift (where genes change randomly) – genetic diversity easily emerges, which in turn creates diverse behavior in individuals. After 500 generations of evolution, about 60% of robots never emitted light near food, but about 10% emitted it there most of the time. And some robots started to emit light near poison, luring in other robots. Some robots were slightly attracted to the blue light, but a third of the robots were very attracted to it, and another third did not respond to it at all.
Experimenters think similar processes occur in nature. When animals move, for example, to get food, they thereby unintentionally signal it to other animals. This creates a conflict of interest, and natural selection will allow those who can suppress or modify this information, e.g. by camouflage, stealth, interference or false signals, to live better. As in the robot experiment, these processes can help to understand and explain a great variety of behavioral strategies in the natural world.
This research suggests that robots may have applications far beyond the technical realm. As one of the researchers who conducted this experiment said, "Robots can be quite useful for studying and better understanding the mechanisms of interaction between living organisms. This is just the beginning, but we are sure that robots will be used for research in biology, psychology or medicine.
Video of 6 robots looking for food (14 Mb)
Schematic video of robots fighting for food (20 Mb)