I am still quite intrigued with the idea of a car driving itself.
So with that in mind, I figured I would outline some thoughts.
For starters, since i don’t have a vehicle outfitted with servos, I would start in the digital world.
Take the video output from a racing video game, and try to analyze it such that it is able to drive the virtual race car.
In more detail what this would entail is a 2 PC setup. One to run the game, and one to analyze the video and control/emulate a steering wheel device.

I don’t know much about vision analysis, so I thought if using something like National Instruments LabView with the Vision package to start off, and see what that can do for me. LabView would be nice since it would lend itself well to controlling the other PC by acting as a steering device.
It would also likely take advantage of multiple cores (so i could get a multi-core CPU and have all the power be utilized) as opposed to trying to write some multithreading code myself (yikes… where do I even start!)
I anticipate that analyzing video may end up needing lot and lots of CPU power. To start I’ll use whatever equipment I have available, but I may end up needing something like a Core 2 Quad with a few gigs of ram. Just a base PC with lots of computing power.

Interesting vision software:
http://www.roborealm.com/index.php
http://www.roborealm.com/links/vision_software.php
http://www.darpa.mil/grandchallenge05/team.html
http://www.arpa.mil/grandchallenge05/TechPapers.html
http://www.darpa.mil/grandchallenge04/tech_papers.html
http://www.google.com/search?q=vision+software+follow+path&ie=utf-8&oe=utf-8&aq=t&rls=org.mozilla:en-US:official&client=firefox-a