Tuesday, October 23, 2007

electric sheep


Video: Just a robot. Not an evolving one. But this one can do a cool trick with a pendulum.


I asked the World's Best Mechanical Engineer to weigh in on the self-evolving robot idea. This is his response:

The You-Tube presentation really whitewashes the real difficulty with self evolving robots. What he showed, and described, were robots that used evolving strategies for movement. But the robot design did not evolve at all. And in reality, the different strategies the robot used were limited to the imagination of the programmer. So the robot was really just an automated test rig for different movement strategies.

Now what would be really interesting would be a self reproducing robot. Though dauntingly difficult, it probably is probably at the fringe of today's technology.

If you had a machine that could mine it's own iron ore, refine it, and build a duplicate of itself, then you would be constructing the conditions for true evolution. Random variations would occur in the manufacturing process. The good variations would survive and replicate, the bad variations would be inoperable and die out. The significant barriers to something like this would be having the onboard ability to make silicon chips, the onboard ability to process some sort of organic materials for electrical insulation, and the onboard ability to build a power supply. The shear size of the machine could be so vast that it could never lumber from place to place to gather the raw materials.

Possibly this could be solved by subdividing the tasks into two machines that work together. Each machine specialized for special tasks. This would reduce the complexity of each machine substantially. Parts of the two machines that serve identical purposes (such as movement) would be identical for ease of manufacturing. You would have effectively introduced sexuality to the machines, where a male and female are needed to reproduce. This allows one machine to gather resources of one type, while another gathers resources of another type, or actually works on the replication process. Maybe this explains sexuality in animals.

The rate of evolution would have to be very slow, because you would want to be sure that at least one successful machine was built each generation. Random variation would probably be sufficient to drive the evolutionary process.

So when you think it through, the conditions for successful evolution of a machine very much mirror the conditions that nature has created for the successful evolution of plants and animals.

The World's Best Mechanical Engineer


See more from the World's Best Mechanical Engineer »

No comments:

Related Posts with Thumbnails