Surely the way that you would eliminate the huge number of variables would be to employ fuzzy logic controllers.
Back when I was studying A.I. I read some of the reports on a project which was using doppler and IR sensors to make a system that would allow cars to autonomously brake and accelerate without the surging effects as the cars react to each other. The control logic with fuzzyification was frightening and I only saw a brief overview of it.
If you consider the level of control that the DARPA challenge vehicles have:
Welcome
Then its possible to simulate a huge number of factors in real-time, possibly even drifting, but it would be far from easy, and you'd probably need a passenger seat made of SHARC arrays to get close to the processing requirements.