8 replies to this topic

[Xcrypt]

Hi,

As a school project I have to make a game as end-project. Now, I would like to focus myself on AI.

Is there anyone with experience in AI who could tell me a couple of games with AI that are really challenging to program, and why?

Thanks.

[fireside]

Maybe you could take a simple game and apply evolutionary learning to it. One thing that sometimes designers forget is they don't want the AI to be too good. It spoils the game, so you could make them find a 3/4 wins for the player or something. The problem there is there isn't any way to compare players, which they tend to like to do.

[Xcrypt]

AI doesn't necessarily have to be restricted to some kind of enemy. I was thinking about making something like http://lionhead.com/...W2/Default.aspx

But I don't want to restrict myself to one thing. Any other ideas ^.^
Just need some collection of ideas right now, because I'm not all that experienced in AI, and don't know what possibilities exist.

[alphadog]

Really challenging? Try coding a system for beating a 5p dan at Go. :sneaky:

[fireside]

Quote

AI doesn't necessarily have to be restricted to some kind of enemy. I was thinking about making something like http://lionhead.com/...W2/Default.aspx

Keep in mind those games were developed by teams of people over a period of years. For an individual project, I would look at indy games for ideas. The Sims has over 1 million lines of ai code in it. Also, doing a copy will always be doing a copy. It's better to use games as a springboard for your own ideas. People will be more impressed if it doesn't look like a bad copy of some popular game and has original ideas in it. I would look at indy games and ask myself how better AI could improve it, or take the other route and think about how some ai routine could best be used in a game to challenge or engage a player.
Also, if it was me doing a school project, I would pick a core idea that could be expanded with extra time. That way I would always have something to turn in. I've seen people turn in vaporware at school, and it's not at all impressive.

[geon]

Any AI is challenging to implement. Most is pretty much impossible to implement convincingly unless you cheat.

I *can* tell you something fun to implement. Try Othello/Reversi with the MinMax algorithm. I'ts pretty simple once you understand it, and can give pretty good results.

[Stainless]

I have always used two types of AI's

The first relies on a search, you come up with a way of evaluating the current situation to produce a score and then run through all possible moves looking for the best.

The best example of this is a alpha beta minimax search.

The second is a layered approach.

You have a top layer that makes strategic decisions, a middle layer that handles planning a way to achieve the strategic aims, and a lower layer that actually handles the actions.

So the top layer says .. "I want to go to the top of that hill"
The middle layer says .. "This is the route I want to take"
The lower layer moves the character along the route and reacts to local events.

Each layer has a set of inputs.

The strategic layer knows about the overall game state, uses things like influence maps to decide what to do.

The middle layer uses things like A* to plan movement, and terrain scanners to work out in detail where to go

The bottom layer takes game events like "you have been shot at" and can override the other layers, passing messages up the stack when events happen.

What might be a good thing for you to start with is writing your own influence map code, your own terrain scanners, A* etc. and then playing with them.

You will be suprised what can happen and where the code can lead.

After all Lemmings started out as a test for graphic artists to see how small they could get a sprite and still animate it.

[Xcrypt]

Well, first things first, I'm going to need pathfinding that's for sure. Is there anyone who knows how pathfinding in modern games are done? Do we still use a 'grid'? Should I use squares or pentagons / X-gons ? A*? Know any decent tutorials? (Already learned the basics so)

[Kenneth Gorking]

You should read this paper by the guy who did the AI in FEAR, which I always found impressive. It has lots of details, and the source is available in the SDK for the game.

http://web.media.mit...n_jeff_fear.pdf