newbie question

[SirYakalot]

I apologise if this is an annoying question. I just joined this incredible community today and I'm a little lost in the vastness of it all!

I'm currently in my final university year and studying AI for games as part of my course. For my final year project I'm going to attempt to create some bots that adapt to how a person plays by attempting to implement an evolutionary algorithm.

I'm fine with the code and the theories, and I know C++, but I have absolutely no idea how to get to a point where i'm coding in visual studio and putting that into a HL2 mod.

How does it work? Is there literally a block of AI code in a HL2 mod that you can see, delete and alter? Or do you have to do it all from scratch?

All i've been able to find so far is how you import new skins, maps etc into a mod, but nothing on how code is implemented.

If anyone could be so kind as to link me to some relevant threads on the forum it would be very much appreciated. I had a look around but there's so much here I didn't know where to start!

Thank you in advance for your help,

I should hopefully be able to start posting some pretty interesting stuff soon!

[The Storm]

I don't have the HL2 SDK but from what I know you should have the complete HL2 source code and also sample code for basic bot that will show you how to start. Look around the sources. I may be wrong tought... Lets hope that someone more expirienced with the HL2 SDK will help.

[SirYakalot]

Actually, perhaps a better question would be:

What would people recommend I use for a final year computing project where I want to experiment with implementing genetic algorithms and neural nets to code some bots that exhibit learning behaviour?

My two main problems are:

I don't want to have to code a whole new game myself as I don't have enough time - so i'd like to be able to just augment the AI of a 3d game with a genetic algorithm as there is only so much 'evolving' a 2d bot can do.

Testing is also a problem, in an ideal world i'd be able to test my learning bots against very hard-coded, predictable enemy bots that I could set to a number of different behaviours to see how they'd learn and evolve.

The reason I thought the source engine would be a good choice is because, in my naivety I thought I could pitch my learning AI bots against the half-life bots. Also the source engine already has a number of built in aspects which would be incredibly useful to not have to code myself - these are things like different body parts taking different amounts of damage, the ability to jump, crouch, pathfinding already built in etc..

however I'm realising that this is nowhere near as easy as I had first imagined.

Are there any frameworks out there that exist for AI testing purposes?