It may be dated, but Judity Dayhoff has a pretty good book called "Neural Networks Architectures." It was one of several primary sources for my thesis work. The mathematical derivation is very lucid and I was able to go straight to code from her formulae. It may have been a bit weak on applications. In particular, I don't recall anything about robotics.

However, I note that nowadays there's a lot of good stuff on neural nets, genetic and evolutionary algorithms, and fuzzy theory archived at NRL.


k