Skip to content


Robot Evolution

Creating robots which have the ability to grow and evolve has been one of the most intriguing aspects of AI research in recent years, as the process of learning new things on one’s own is a definite sign of intelligence. Scientists at the Laboratory of Intelligent Systems in the Ecole Polytechnique Fédérale of Lausanne, Switzerland, have been working on miniature prototypes which display exactly those qualities. Their robots have been able to work together and share information when navigating through a maze, and have even mastered their skills in a type of predator-prey relationship—the implications of which, I think, speak for themselves.

The robots are specially designed with random variations that can change through virtual mutation after every experiment. Much like the real process of evolution and selective breeding, the bots that had been more successful in the tests had their gene codes passed on to the next generation, while the ones that failed did not. This neural network allows for each new generation of robots to be an improvement on the last one.

One truly remarkable result came after a test designed to create a predator-pray relationship. One of the bots was the “hunter”, which had better eyesight and could see its “prey” from further away, while the bot being chased had superior speed. At first the robots just shifted towards and away from each other without much apparent thought, but after 125 generations of selective breeding, things changed. The predator bot developed the brains to stalk it’s pray from the sides and hide behind walls, where its enemy’s sensors would not be able to detect it. The prey, however, learned to rotate in its place and scan all sides, effectively nullifying the hunter’s strategy. When it increased its maximum speed and got too fast for the hunter, the latter started setting traps and lying in wait.

Other experiments involved the robots working together to push tokens of various sizes around an area. Some bots elected to push smaller tokens on their own, but those that were awarded the most points and the most offspring were the robots that teamed up to push the larger objects.

Where all this is leading is up for debate, but it is fascinating stuff nonetheless. In the meantime, enjoy the below video of a robot evolving to navigate a maze.

Scientific Memory Games: Improve your cognitive abilities.

Go to Source

Posted in AI & Robotics.