Couldnt you basically make a script language that had as a base all the basic actions that an NPC could do. Then you make action classes, for example the Attack Nearest Monster class would have an action that would target the nearest monster, the actions too move too it and then the attack action. Then in some action classes you leave possibilities, like instead of writing every possibility you have a Rise in Rank class with subclasses like Kill superior, work hard, bribe someone. The subclass could be chosen through your attribute system. For the King and his brother scenario, the Prince wants power, so the Rise in Rank action (that is associated with his personality type) is activated and because of his attributes he choses too murder superior, the superior is the king, so he has chosen to murder the king. Then he has subclasses in murder superior, all forms of killing a superior, including assasination. Because of his personality type he again chooses assasination.

Basically, it would be like writing up a whole bunch of scripts and letting the NPC's choose which ones server them best based on their AI type and perhaps also previous events, not sure how that would work in though.

Edited by - KingMolson on January 23, 2002 11:05:11 PM

I find it VERY amusing that people love to re-invent the wheel many times over! Might I draw your attention to the AI field known as planning ??? In particular, if you want to implement a system of desires (which is really a preference scale for outcomes of goals) then consider decision-theoretic planning .

quote: Original post by Nazrix
Yes I went through that same dilema. Basically it's my understanding that an agent or NPC can only know something if:

A) we (the desingers/programmers) tell him
or
B) the NPC learns from some kind of memory/learning type of AI

So far with my AI I have been going with the idea of telling the NPC by giving him the info to make a decision. I know it's not as elegant as an AI mechanism where the NPC is really making a total decision, but I am not sure we can do better for real-time apps like games.

You can do better much than that. Implementing decision-theoretic planning in a game would be a fairly trivial matter since it's a "toy world" problem.

References: There's an excelent book by Thomas Dean called, "Planning and Control". It has much more than you need for a game AI, but you should find it useful. Alternatively, check out his article, "Decision-Theoretic Planning and Markov Decision Processes", which you can find on the web. There are plenty of other authors who work in the field. If you check out www.citeseer.com you should be able to find related work by seeing who has referenced Tom Dean's articles and who he has referenced.

If you have specific questions, I will be happy to answer them.

Good luck,

Timkin



Edited by - Timkin on January 24, 2002 9:50:35 PM

quote: I find it VERY amusing that people love to re-invent the wheel many times over!

There are better wheels, and less costly wheels. In fact, I''m pretty sure new wheels are being invented even as we speak. Should we discourage the practice ?

What I mean to say is : planning in games is far from perfect so it doesn''t hurt to think it through every now and then.

Thanks Nazrix, that stuff was really good reading.