Greetings. I just wanted to post my project and get feedback from other, more experienced, programmers.

I've been working on an open-source game called Armagetron. It's basically the light cycle game from the movie Tron. The basis for this game is that you are in a small grid arena and as your move; your trail leaves behind a wall. This is continuous until either you or your opponent hits a wall. Winner is the one that survives the longest.

Awesome game, however I find the AI.... shall we say a bit lacking. They usually trap themselves and from the code it looks more like the program is just checking for an opponent wall. There is very little in the way of self-preservation, learning, or any form of strategy. If you watch the AI play itself, usually the two bots will box themselves in and die.

So, my plan so far is to implement a rule-based system into the game with weights on certain variables. I don't think that I'll implement learning or a neural network at this time because I believe that they'd be too slow to react in the game. What do you think? Any suggestions or criticisms are welcome.

My page with the proposal and the game: http://www.nthelp.com/james/AI.asp

Good RisingShadow,

My name is Ed. Your game looks to me filled with humor. I like your game. So stupid. The bots will box themselves in and die. End game. What now. Is this English?

I once coded a game and the first thing the computer said. You are devoured by bats!!!

And this was so humorous.

Maybe someone knows that game with a pit and other nasty things, and the goal was to find the structures of the corridors that connected some 30 rooms, I guess. You smelled one room in advance. A funny but stupid game. But the typo in the source caused humor:

Imagine, n o k e y h i t ... and already devoured by bats!!!

...Swallowed. Setting shadows.

The movie Tron. Trapped, that's bad, and luck, it's a game. So atmosphere, isn't it. Some nice graphics, sounds maybe and... wow.

Why don't you add self-preservation, learning, or forms of strategy?

Strategy you learn by acting, self-preservation is a strategy also so again acting and then learning.

So all solved when you can act and learn, apply and learn more, maybe much more. Our brain is large enough isn't it.

I think your making a really nice game. What are setting shadows? Maybe also a version with the opposite when the bots are safe within walls but they can eat walls as well, or what about some switch between versions. Only fun I think.

Is there a winning strategy? I hope not. But does someone know that?

You can always start with a strange beginning of some walls, not necessarily connected. Maybe you could break then a winning strategy. You can always add some tiny randomness to get unpredictability. What do you think?

Ed van der Meulen

Ed said:
"Maybe someone knows that game with a pit and other nasty things, and the goal was to find the structures of the corridors that connected some 30 rooms, I guess. You smelled one room in advance. A funny but stupid game. But the typo in the source caused humor"

I believe that was called Wumpus.

Thanks for the somewhat fragemented reply. I don't totally understand what you were saying, but yes, self preservation is a strategy. However I'm doing this as a project for my Integrated Research in AI class. Which means the AI I'm going to write will have to be a bit more generalizable. I'm thinking that with a weighted rule-based AI it could also be easily converted into various other pathfinder programs.

Hello RisingShadow

I talked about Wumpus, I think you are right and you wanted:

to add self-preservation, learning, or forms of strategy. I answered

Strategy you learn by acting, self-preservation is a strategy also so again acting and then learning.

And so it's an integration of acting, learning an applying knowledge.
Learning and in the same time applying and extending the knowledge, like humans do.

I was a little bit playful, because the memories of Wumpus. I added some new ideas for your game as well, inverting the rules, switch between them some time and more things.

You are path finding in 2 dim. 3 dim is more difficult.

Suppose now you are one of the bots.

1. How would you react in general
2. what knowledge do you use then
3. with what knowledge do you extend
4. how go you integrate the knowledge
5. do you act and learn in the same time?

How can you simulate this.

When we look at your game then in humans there's no much difference between learning and functioning.

I think in our actual life that learning functions in a more protective environment.

I know this is a little top-down reasoning, from questions to realization, but maybe it gives you ideas.

Ed van der Meulen