The coolest thing I've seen this week is the concept of 'motor babbling' - that is, the way a robot 'babbles' (just like a baby) as a way of using neural networks to define and create a map of its sensory-motor environment. I found the concept - and a detailed description of how it works - in this paper, one of many I read following up the postings from the Robot Cub. Ross Dawson reports, "We are at the cusp of a new phase of robotics, where some of what has been promised to us for decades will come to fruition. An example of this is the iCub, a humanoid baby robot that can learn, emulating human cognition and development. This is the field of ‘developmental robotics': creating robots that can learn and develop their capabilities over time." The best learning theory, I say, is one that actually results in learning. Don't miss the videos on Dawson's post.