So sad: Tribe used to be full of interesting types who'd likely have some input here... but now? Yutzville.

MIT used to have some great white papers on such topics.
I'll have to see if I still have copies of them.
One of the most promising involved a code structure methodology that, if memory serves allowed complex behaviors to be created from the interaction of a network of simple asynchronous functions that accept a standard stimulii record(much like a GUI handles user input events) and can either forward the input upstream (unprocessed), create new stimulii, until it finally terminated in a command to an actuator, which may also issue stimulii records...

The methodology lended itself well to small C implemantations, and to simulations written in higher level languages. There were even a few examples of using the methods in PIC assembly language.

It also has good scaling properties because the stimulii event records lend themselves to distribution over RANs(Robot Area Networks) where there may be a few large computational units, with large internal arrays of stimulii operators, or a fabric of small computational units each containing only a few operators.

If the standard operators, task management, and other common elements were factored out into a small set of libraries it would be possible to share various types of common behavior patterns across many different types of robots enabling code reuse.

The approach tends to favor development of networks of simple operators, rather than large blocks of procedural code. The resulting behavioral networks are easy to create, and test, and ar often very code efficent!

=B-)