AI Continuity

It''s all in the heuristics.
You''ve got some heuristics right and they tell each AI agent what to do when they''re running around the gaming world. If you''ve got a set of heuristics to decide what the individual should do in a situation based on their current health, their stats and the objects around they could attack then you really shouldn''t be weighting them so that an individual thinks "I''m well hard with health 100%, I can attack that castle on my own. Ouch, I got hit and I''m now on 90% health, I best run away".
This just isn''t a sensible heuristic for an agent to have.
Make sure they pick targets they can overcome alone or as part of an army and make sure they''re targets that they think they can take with x% less health. Also, if you don''t wan''t an agent to change it''s mind you can either make them update their ''general'' decision making state every x seconds rather than each frame (which, let''s face it, is unnecessary) or you could put a minimum limit on health they have to lose before they can change their mind about attacking.

Just give them simple criteria for what they can attack in a given state, a delay before they can reasses the situation again and limit the circumstances before they can change their minds.

You could even, though everyone seems against it, let them run battles against themselves and adjust the heuristics towards an optimal solution using GA''s.

That help or hinder? ;-)

Mike

Do what people do. They evaluate their own actions. How many times have you caught yourself waffling back and forth on a decision and then realized it was silly and you were accomplishing nothing?

Your agent needs to be aware of what it did two seconds ago. In general, if an agent keeps changing its mind, it must be because either it has insufficient information, in which case it should stand back and take the cautious route, or it should create a goal to obtain more information to better solve the problem.