"Robot generates several self-models that match sensor data collected while performing previous actions. It does not know which model is correct."
Thats nice and all, but there could be thousands of possible connections, many of them aren't correct, and many will appear to be correct with the data it has. Also, it leaves the proccess of actually picking "self-models" pretty vague. Does it pick them out of a hat? Thats not usually a good way to solve complex problems. Still, I like the concept, I just wish it would be a little less vague.