MIT Technology Review

Evolution has worked pretty well for biological systems, so why not apply it to the systems that control robots?

North Carolina State University and University of Utah researchers have advanced evolutionary robot controllers by combining artificial neural networks and teams of real mobile robots to demonstrate that the behavior necessary to play Capture the Flag can be evolved in a simulation.

The method could be used to develop components of control systems used in autonomous robots that carry out tasks like clearing minefields or searching for heat sources in a collapsed building. Eventually, the method could be applied to robots that must learn to operate in environments that humans don’t understand well.

The capture-the-flag learning behavior evolved in a computer simulation. The researchers’ randomly generated a large population of neural networks, then organized individual neural networks into teams of simulated robots that played tournaments of games against each other.

After each tournament, the losing neural networks were deleted from the population, and the winning networks were duplicated, altered slightly, and returned to the population. After several hundred generations, the neural networks had evolved well enough to play the game competently and were transferred into real robots for testing in a real environment.

The method could be used to develop practical control system components in 3 to 6 years; it will take 10 to 50 years to develop evolutionary robots that can automatically adapt to unknown environments, according to the researchers.