Robotic Evolution Update

As regular readers know, we follow robotics development on the blog. The video above shows an Apprenticeship Learning AI based robot flying aerobatics with a radio controlled helicopter. This also represents the current state of the art in outboard computing controlling another potential robot (in this case the helicopter)

For about 20 years various types of AI and EA (Evolutionary Algorithms) have been used to self-teach robots, but until recently the computation needed meant they had to be outboard of the robotic vehicle unless it was quite large. But, one of the impacts of Moore's law has meant that now computer miniaturisation means they can run the controllers onboard (especially with EAs, which usually generate quite compact programs), so the robots can evolve their intelligence in real time (though starting off with a flying robot is not advised - most learn to walk before they fly ). This starts an entirely new chapter in robotic design.

Another interesting evolution is this fascinating talk on TED by Hod Lipson, this one about using EAs for robots to build themselves. For about 20 years people have been using EAs to let robots work out their own way of controlling themselves, but using EAs for them to design themselves is much newer.

The future of robotic evolution increasingly looks like it will be self evolution - software robots can already self replicate, it will be interesting to see how hardware self replication evolves.