Half an Idea

I was interested to see that someone has created virtual robots and got them to teach themselves to move through an atifical intellegence programming technique called a neural net. The idea is that you don't tell the robot anything about his environment or how to move, but you give him the "will" to stand upright and to walk by scoring any movement that approximates to those targets more highly than any movements that doesn't and allow the robot to remember the moves that gave the best result and to then try variants on those moves. See the article in the Daily Telegraph.

If you follow this link you will see the robots in all their glory (one day I will work out how to embed this in the blog)

This is pretty much how a baby learns to walk, except the baby's scoring is based on how much falling over hurts and how close to mimicing his parents and siblings she has come.

Brilliant!

I am sad to say that I thought about using neural nets to programme robots to walk several years ago, but couldn't work out how to build a robot in the first place, so I shelved the idea. I didn't think of building a virtual robot and I am not that I am sure that I could do that either, but it just goes to show me that half an idea is still an idea...