I've never heard of any commercial games using fuzzy logic before. But the utility system mentioned above is broadly equivalent to fuzzy logic in most ways, in that as it generates a continuous value for each behavior based on several distributions of input values. In fact, from a designer's point of view, it's potentially almost identical - it's just the underlying mathematics that are likely to differ.

The main difference is that in FL, you define your defuzzification thresholds on an individual input. Those areas map input to a selected output to some extent. It is difficult to map multiple inputs into a defuzzification zone, but it can be done. (I wrote an article about a way to do this with 2+ axes for AIGPG 4.) When you get into n-dimensional space, the model is a bitch and a half.

That's why I came up with the IAUS because it is so much easier to combine an arbitrary number of inputs into one scoring function.

The other main difference is that FL provides you an output for that input whereas the IAUS (and other utility systems) can simply provide a score. That way, it is easier to weigh dozens or hundreds of actions against each other by comparing scores rather than the authoritative model that FL uses ("this input is N, therefore our action is A").

These evaluation functions (fuzzy or otherwise) are still just lower level tools to be employed by a higher level decision making (and action guiding) framework. Solution sets are modal as regards to the situational factors and can be severely different and even inverted for the particular problem being solved. The process goes : Classifying the situation, then looking for solutions to multiple potential strategies (which then make use of specific analysis specific to each goal type) and then carrying out the execution of the strategy (tactical steps which adjust along their progress).

So lots of evaluatiion functions are needed for the different specific solution proposals. Flexibility to use dfferent approaches - some may have many factors requiring a fuzzylogic like approach, while others have very few relevant factors and simpler logic can be employed. Those would be used themselves within option Searches (like a targeting scan) which have their own parameters relevant to the goal its being carried out for (like distance or terrain considerations).

Priorities (part of the decision metrics) shift in non-linear ways depending on the condition (ie- you are unhurt, hurt a little, hurt alot, critically hurt -- which drasticly shifts the importance of certain goals and abilities to attain goals. Yes fuzzy type logic handles that but decisions controlling whole goal sets are best controlled at a high level, and that logic sits ABOVE at the goal selection level (which controls all the lower level processes).

Point of all this is that for something more complicated than a oldschool game object, with its flat instant decision making, it requires a great deal more layering and complexity (and all the human work to tune it all into a cohesive system).