Don't get me started on intelligent algorithms!! I'm probably the only guy on this forum who thinks intelligent control in embedded systems is practically feasible (although for highly specialized applications)! That's coz most people get by pretty well with conventional methods. There's plenty of algorithms to choose from, depending on the sort of problems and unpredicateble situations your robot (or process) may encounter.

The normal method is to have some action defined for ALL contingencies. The difficult thing here is to think deep and try to predict all these contingencies. This is what happens in most embedded control systems, and most practical systems really don't encounter many unpredicted scenarios.

However in case of complex and/or inconsistent systems, you need something more... the algorithm should be able "learn" from its experiences in unpredictable situations, so as to be able to better handle them later. This calls for Aritificial Intelligence (AI)!

However before going full-fledged AI, you could try Fuzzy Logic. Fuzzy logic is a powerful tool (rather than an algorithm) you could use to determine which expression(s) best describes the current state. You can then take a control action based on the expression(s), which is defined linguistically rather than mathematically. Fuzzy Logic can allow a system to react even when faced with undefined conditions. It is the easiest thing to learn, and very applicable to embedded control systems. As far as I know, it isnt very popular in the US, but has caught on really well in Europe and Japan.

If you wanna get really advanced you could look into AI algorithms. I'd guess Neural Networks is the most applicable to embedded control systems. Fuzzy Logic is generally used as a learning tool in AI algorithms. From the hardware point of view, you need some sort of non-volatile memory to store wotever your algorithm has "learned", maybe an NV-RAM. However the procesor must not neccesarily be very powerful, since many types of AI can be implemented in pretty small and/or simple programs.

You could also try stochastic and statistical analysis to tune your control system for stability.

kundi