Hey everyone, I need a little help with this. I should be starting my Diplom project soon and I know I want to do it on AI and preferably something along the lines of player prediction and/or goal setting but I don't really know what to do there... In case you don't know: The Diplom project is the graduation project after a course of studies of four years here in Germany. Now, I have decent knowledge in the basics of AI, I know the A* algorithm and implemented it, I know the alpha/beta cut and used it for a Tic-Tac-Toe game in Java where you simply couldn't win against the AI, I know separate and conquer, strips planning, neural nets, basien nets (I hope I spelled that right though ;) ), genetic algorithms, a priori... The basics of all these subjects. Unfortunately there is so much out there that you could do as a project even when it just comes to player prediction or goal setting that I am a bit at a loss. What would you find interesting? Really, any ideas on projects would help me at this stage, since I just can't come up with any good idea for a project and actually know how to do it... I would be very grateful for any kind of help.

How about a flat (2d) top down real time strategy-type a* algorithm.

Basically you have lakes and trees that need to be navigated, the thing that is moving has a certain size. What is generated is vector information, if you tell the vehicle to go to the other side of the wall you don't want it to go directly to the wall and then move around. So the vehicle moves in a vector towards around the end of the wall. The result of the algorithm would be vector information.

Extensions are: Taking into account turning arcs (and reversing) (for vehicles), terrain cost (mud etc), hot spots (enemy fire area that if possible shouldn't be entered) and if possible group dynamics (a group of units should arrange themselves in formation).

All these things that are used in rts games

The good think would be the graphics can be easy - just rotated rectangles for the vehicle, bitmap for the terrain / buildings.

Thanks for that idea. I'll be playing around with it a little. :)

I've tried finding Diplom projects on AI in computer games online, too - but so far I haven't found anything. Neither finished ones, nor any proposals... Funny, I thought it would be easier to find something.

you can check out for J.M.P. van Waveren's thesis on quake3 aerna, as a reference for your project.

http://www.idsoftware.com/home/jan/q3abotai/Q3ABotAI_15.pdf

Quote: Original post by Kassandra
Thanks for that idea. I'll be playing around with it a little. :)

I've tried finding Diplom projects on AI in computer games online, too - but so far I haven't found anything. Neither finished ones, nor any proposals... Funny, I thought it would be easier to find something.

Add to that suggestion having a bunch of moving objects all over the map that cause the paths to have to be constantly revised.

A clever scheme to lower the amount of work done -- instead of a 'brute force' recaculating the path every step.

Possibly being adaptive to adjust the reevaluation frequency/extent to react to the situation....

Quote: Original post by Anonymous Poster
Add to that suggestion having a bunch of moving objects all over the map that cause the paths to have to be constantly revised...

Good idea. The nice thing about this idea is that you could (if done right) add more ideas if you have the time. For example you could also add a heightmap to make the algorithm 3d (take into account line of sight, slope of terrain, etc.) If you wanted to go even further you could implement a simple resource based rts game, and program the 'global' ai for the computer players. Resource management, army tactics, etc.

Just head toward one of your AI Profs or one of their assistants.
They should help you (that's what they getting paid for).

I did my diploma project on AI too, and I worked out the task with the assistant of my AI professor.

edit: Just implementing a* or something like that will be not enough as I think. You should come up with something new, like "what algorithm is best for problem X" or "how can X be improved asuming conditions Y".
The diploma project is not coding, but it is research on some problems. Some people just code out their project to prove their theoretical results in a practical way.

If you decide to do anything on GA or GP feel free to email me (ceo at grexengine.com) - I did my uni dissertation on "Evolving players for Pacman using Genetic Programming".

Thanks everyone. I've done a lot of looking around lately and got some more ideas on what I could do... I'll have to think about it a while longer and actually write a couple of things down, take them back to the professor I want to do this with and see what he thinks about it and what he thinks will be possible.

The reason, I am waiting to go back to my AI professor is that I had asked him about personality modeling in computer games and what I could do for a Diplom project in that field a little while ago. The result was that he kept talking for about half an hour, partly on the actual subject I asked about, and told me about a hundred different things I could do in AI and that game AI has "mostly been done" as he said (which I disagree with) - and at the end of this I didn't know any more than I did when I started out. He's a nice guy but when he's really enthusiastic about something - like AI - he talks your ear off beyond the point of it being helpful. [wink]

So, I decided to go back to him, once I've had a couple of much more concrete ideas.

Again, thank you all for your help.