AI Line Helper

Apps AI Line Helper 2022.03.08

Login or Register an account to download this content
There are still limitations to the AI in general so do remember that. But they key thing to try to avoid is 'pinching' the main center AI line up against the edge. AI see the edge as a 'wall' so if the line squeezes up against the edge line too much they can freak out. In some cases it still may be required to get the line in the editor so you can fine tune certain areas. Also when recording you sometimes have to make a compromise between an ideal line and one that works. You may also have to use ai_hints in areas like that to slow them up a bit more entering that area.
 
Now that we have this tool working (hopefully) I'm guessing we'll need to figure out some best practices for which vehicles to use (small with tight turning radius) to drive the lines and how the lines should help the AI through various types of turns and road conditions. @LilSki had mentioned keeping the line at least 1 foot from any wall. I was working on Hockenheimring AI last night and they are bunching up at the tight chicane especially on lap 1, so still need to figure that one out.
The Abarth 595 at 30km/h makes a good tool I think. I'll try a new Lucca Ring line next (the current borders reach outside some tire barriers and form crevices around the roundabouts, right side also goes right across the wobble curb before the pits)
 
I like to record it with the Honda S800 by A3DR, because it’s a cabriolet and you can see the road better (I tend to record in chase perspective, that might explain that).

One question: others can download side_r and side_l, install it to the correct track and it works for them, too?
 
Guess its a good idea to bundle these into the upload if we publish updated AI line here, so that people can applicate them to their fast_lane file.

Exactly. Now with the new app I’m reworking Bahrain as the track limits there are flawed in some parts.
 
Exactly. Now with the new app I’m reworking Bahrain as the track limits there are flawed in some parts.
I did new lines for Lucca Ring last night (~8 minutes a lap at 30 km/h :roflmao:), which work pretty good so far. Cars still bump into the walls sometimes, but with a very mixed fast AI grid I only had a single car retiring after a crash during an overtake, in a 20 lap race. Before the changes half the field kissed a tree in the first few laps.

I may have a go at Tiago Lima's Cicuit de La Sarthe next. Any other ideas for tracks that need work?
 
  • Like
Reactions: F_B
As an fyi, the fast-lane files have the track boundaries "baked in", so you should not need to distribute the side files unless you want to help other folks make AI lines with them.
 
I did new lines for Lucca Ring last night (~8 minutes a lap at 30 km/h :roflmao:), which work pretty good so far. Cars still bump into the walls sometimes, but with a very mixed fast AI grid I only had a single car retiring after a crash during an overtake, in a 20 lap race. Before the changes half the field kissed a tree in the first few laps.

I may have a go at Tiago Lima's Cicuit de La Sarthe next. Any other ideas for tracks that need work?

Sounds great, Lucca is worth it, such a well made track.
Road America still has the zigzag issue after the Sargento bridge, perhaps it’s related to track limits. Will give that one a go next time. I think there are plenty of tracks which could need better track limits in order for the AI to work better.
 
As an fyi, the fast-lane files have the track boundaries "baked in", so you should not need to distribute the side files unless you want to help other folks make AI lines with them.

They are automatically baked in to already existing fast-lanes? Or do I have to record a new fast-lane after I made track boundaries in order for them to work?
 
Any other ideas for tracks that need work?

Hockenheimring and Mexico 88 are on my short list. If you load the dev tool for making fast-line files and see odd red lines around the track borders then creating a fast-line using side lines should help AI be less erratic.
 
They are automatically baked in to already existing fast-lanes? Or do I have to record a new fast-lane after I made track boundaries in order for them to work?

When the AI fast-line files are recorded they store whatever boundaries were being used at the time. There is no way of knowing if the person recording the ai lines were using side lines or just surface.ini definitions.
 
When the AI fast-line files are recorded they store whatever boundaries were being used at the time. There is no way of knowing if the person recording the ai lines were using side lines or just surface.ini definitions.

Would you advise me to record a new fastlane after I have defined new track limits? That’s what I still don’t get: if old fastlanes (before using your app) will get the information of track limits (ergo ’baked in’) or just a new fastlane.
 
Would you advise me to record a new fastlane after I have defined new track limits? That’s what I still don’t get: if old fastlanes (before using your app) will get the information of track limits (ergo ’baked in’) or just a new fastlane.
You don't need to record a new line. That is sort of the point of all this. If you load up and get the yellow lines then it has baked the data into the current fast lane file.
 
  • Like
Reactions: F_B
You don't need to record a new line. That is sort of the point of all this. If you load up and get the yellow lines then it has baked the data into the current fast lane file.

So you could delete the side files because it’s now baked in the fastlane? Just trying to understand.
 
You don't need to record a new line. That is sort of the point of all this. If you load up and get the yellow lines then it has baked the data into the current fast lane file.
I was under the impression that the yellow lines indicate that side_l.csv and side_r.csv files are being used to determine track limits, which you can then use to record a new AI fast-line, and recording new side lines would not affect the track boundaries "baked in" to any currently existing AI fast-lines. I can always be wrong.
 
Back
Top