Skip to main content

Robots Simulate Evolution: The End is Nigh

Once robots start evolving on their own, we won't be able to stop them from assimilating us, even with Patrick Stewart on our side. If previous reports on robots using genetic algorithms weren't warning enough, it's best to take heed now. Josh Bongard from the University of Vermont is helping the robots evolve, and this time around, they're not just dinky little rods and cylinders.

Unlike the micro-Borgs from Frauenhofer Institute, Bongard's robots start their evolution in the digital realm. Each of his virtual 'bots were told to seek out a source of light, and figure out the best leg configuration for the job. One had no legs, another started off with four lizardlike legs, while a third had upright legs. The first two were given the genetic algorithm, while the latter served as control.

When all robots evolved to the point that they could walk upright, Bongard took their respective algorithms and applied it to a real-world robot. What Bongard discovered was that the real-world stand-in functioned better when using the algorithms from the evolved machines, as opposed to the unevolved control.

This, and other experiments like it, will help ease engineering problems by developing robots that can adapt to their situation instead of building them from scratch. So remember folks, once we're pickling in the human battery pods or serving aboard a Borg cube, we have engineers like Bongard to thank.

[source: New Scientist]