Objectively Measure a Custom Vehicle Setup – By AI?

[wnx_echo0010]

Has anyone tried to use the AI as an objective measure of a custom vehicle setup? The basic idea is to enter your customized car into the One Make Race option under the Practice section. You could then compare your results with other vehicles or other tunes of the same vehicle. Read on for more thoughts.

This idea came up as I was designing a database application for GT5 that could recommend a particular vehicle / setup for any given event, A-Spec, B-Spec or otherwise. I know a lot of people are putting effort in this direction but I haven’t yet seen anyone try to incorporate all of the possible tuning options.

One idea I had was a tune review template that would make it easy for anyone to try out a custom setup and post up their particular results. I would ask them to post their best times on be three standard tracks with that particular vehicle setup. Once you got enough data you could post some really interesting analysis but one very cool thing you could do is actually recommend a custom tune to someone based on the event, the driver’s skill level, and a difficulty setting.

I was thinking about how to collect all of this data and I began to wonder if there was an easier way. Yesterday I took a custom tuned car into the Practice area and entered it into the one make race. I began to wonder if you could use the AI results of this race as a sort of ‘standard’ benchmark on a particular vehicle tune track combination.

Using the AI is of course problematic. I don’t know if the AI in a one make race would actually use my custom setup or not. Even if they did, would they push the car hard enough to really bring some of these fine points into play or would the deciding factor basically come down to the ration of power to weight? If I set my brake balance to favor over-steer I doubt the AI would take advantage of this in a tight corner. (Honestly I barely know enough about tuning and racing to even come up with a valid example. I am really a database guy who really likes this game.)

Having nearly talked myself out of using this idea at all I still wonder if anyone else tried to use the AI as a sort of objective measure to compare one vehicle setup against another.

 

[BlueShift]

The problem may be what I'm guessing : AI use driving aids. Mainly squid recovery (that's why players can still hit the AI cars and they don't quit the road)...

 

[wnx_echo0010]

The problem may be what I'm guessing : AI use driving aids. Mainly squid recovery (that's why players can still hit the AI cars and they don't quit the road)...

Interesting. If the AI used skid recovery force this sort of benchmark test might be utterly useless.

If I understand the aids correctly they generally modify the driving inputs of a real driver so that you can’t make really bad mistakes. If you punch the gas before you are safely out of an exit then it would modify the input you just gave it so that you would still maintain grip. It would tone down your acceleration command and give as much acceleration as it can. Normally this would cause understeer and send you into the grass any way. If I am right AI don’t really need these aids since they are already programmed to keep their inputs well within the ‘grip zone.’ If they do end up in the grass it is typically because they were forced out of their anticipated driving line usually by the one massive unknowable variable in the equation, namely the human driver.

I think (I need verification) that SRF is different in that it momentarily changes the physics to introduce grip after a skid has begun.

As a simple test you could take a badly tuned car into a single make race and see how it affects the AI. For example you could terribly upset the brake balance in a very heavy / high powered car and take it on a track with wicked corners. Based on the results you could make some basic assumptions. If the AI has no problems with the unbalanced setup then they either do not use your setup or the SRF option is in play. If the AI has huge oversteer problems and spins into the wall constantly you know that they are using your setup and the SRF option is basically not a factor. If the AI is posting much slower lap times but is basically maintaining grip the whole time then you can assume that the AI is using your setup and is going much slower to compensate. SRF may or may not be a factor but you can assume the AI does not lean on it under normal circumstances.

I am still skeptical that the AI will effectively measure a setup that has been intentionally tuned for drifting or loose grip.