Hi, I have to come up with an AI project idea for a course on AI in games that I'm taking. So, I'm looking for a simple idea for a game (or a simulation) that would visualy demonstrate some AI algorithms and techniques. Obviously, I want to emphasis the AI as much as possible so I need it to be easy to visualize graphically. I'll appreciate any ideas... Thanks (By the way, I'm planning to code the project in XNA, so if you have any tips on that too...)

I'd personally be interested in chasing, following, escaping and group behavior

Demonstrating a non-scripted response to environmental events is a good project for a single unit at undergraduate level.

Thanks, any specific ideas?? such as ai in a mortal kombat type game?

What year level is the project (I'm presuming it's part of an undergraduate college/university degree program)?

I get my 3rd year students to design an intelligent pilot for a space-based game I wrote. They have a large number of small problems to deal with: limited resources, limited health, environmental damage sources, nonlinear flight control caused by black holes, smart enemies. They are directed to create an event-driven agent that can react to sensor information, but also plan in order to solve goals. They are given assessable tasks, such as: 'find an asteroid field and mine it'; or 'destroy a Combat Drone'). How they solve these is up to them. One catch though for them: there's no shooting in the game so destroying enemies takes ingenuity!

If you're a first or second year student you should limit yourself to something simpler. For example, write an agent to control a moving character/vehicle that includes both path planning and dynamic avoidance of obstacles that block the path. This is a fundamental problem in game AI and having a solution to this in your portfolio that you later show to prospective employers will improve your chances for employment.

Cheers,

Timkin

Maybe the old cellular automata flock of birds/school of fish system. There's been research conducted suggesting that fish can only perceive a limited number of any given thing (for the fish in the study, it was 4). You could play around with that setting and see the results of perception dynamics.

For example, a school of fish could go in one direction by each fish converging his vector with the vectors of the four nearest fish. However, when the predator enters the fish's field of view, it takes away from its net perception, meaning the school would not only split into multiple groups, as current algorithms show, but exponentially decrease in cohesion.

There are plenty of variants on this theme.

Thanks for the ideas so far!
Any more ideas?

How about a turn-based computer v computer squad game within a small arena where each squad member says what action they are taking and why?
Simpler task is a Battle-Tank scenario where they each have different attributes and learn to detect kill-zones while using FSM to select a steering behaviour (pursue, attack, evade or whatever). Each displays it's current mode or changes colour depending on the situation.