Expert AI for Stock Cars

10 comments on “Expert AI for Stock Cars”

1. Ato is referring to -gazzasaiinfo (include -dev and -gazzasaicar too).
   I won’t doubt you’ve studied her documentation, but what MadRRacerTurbo mentioned happens because of the AI nodes. As you’ve mentioned, AI is static, so they do what the actual node type tell them, but they do it immediately and they do it with their AI Setup values, BUT as good as they are in synch with their rest of the parameters’ values.
   Let’s say, that her AI setup is well built, the whole parameters. If the track has less suitable nodes, raceline, surface for a given car, it will fail, despite the parameters are ‘perfect’.
   And the reason is, because the AI static. They do what they do with all the information they get from the mentioned fields, and nothing else.
   So again, tl;dr: stock and most of custom tracks’ AI nodes are not well built.

   0

2. Sure! If I may know though, how do I find the yellow circle near or around my car? I do use those parameters but I can’t see anything like it. Is it under one of edit modes?

   0

3. Edit: Uhm, the other way around: a little lax on ice, a little tight on tarmac. So I still think that the problems you describe happen more at high speeds and on slippery surfaces like grass. The same correction on tarmac is less effective on grass. And less effective at higher speeds (due to movement inertia) than at lower speeds. Humans compensate dynamically – start compensating earlier if at higher speeds, but RVGL’s car AI is static – same correction at different speeds.
   I’ll shut up now… Go do the test and tell us what you saw :)

   Greetz,
   Ato.

   2

4. Do this test: run the game with the switches -dev -gazzasaicar, try different cars on different tracks, and monitor how “your” car’s yellow targeting circle follows the tracks’ AI nodes and how the car’s AI tries to follow that circle. Then come back and tell which cars and where exactly fail at following that circle. (I tried the pack and didn’t see a problem, they follow inhumanely precisely the tracks’ AI nodes, given their speed and surface properties).
   For clarity – RVGL’s car AI is coded as a static function, meaning that it reacts the same on all surfaces. It applies the same correction thresholds set in parameters.txt whether it is currently on tarmac, grass or ice. This means the more difficult cars had to be tuned to a certain “middle point” – a little lax on tarmac, a little tight on ice, because the car AI basically corrects on _absolute_ values of side slide speed, not dynamic percentages (the latter would have been dynamic car AI, which it sadly isn’t).

   P.S. Nerd mode – yeah, I’ve studied her work closely. I know the ins and outs.

   Greetz,
   Ato

   1

5. Read my reply again. Even if the AI is good, its performance highly depends on the nodes.

   0

6. Thank you. I will try the AI with catch-up disabled.

   Edit: Disabling catchup in profile’s .ini does not change anything. AI still steers before turning much too late. They keep driving straight almost until they hit the wall or another obstacle.

   0

7. Just in case: these were created for Simulation mode and no CatchUp. CatchUp messes with the bot cars’ top speed and then the tuned AI is not optimal anymore.

   Greetz,
   Ato

   1

8. They have this issue on all stock and custom tracks I have so far though. Examples: SuperMarket 2, Game Room 1, Museum 2. They always start to turn very late, no matter what.

   0

9. AI works connected with AI nodes. If any of this two fails to work properly, AI won’t be decent, it will struggle to complete tracks.

   0

10. The only issue seems to be that the cars often take waay too long before starting to turn with those new parameters, which causes quite some trouble for AI at some tracks.

    0