- 351
- United Kingdom
Which is the best way to record an AI line? With a fast car or a slow car? I'm aiming to have races with a wide range of cars from the inter-war period (1919-1939), and their performance level varies greatly. So in theory I could record an AI line with something really slow (like Gary Patterson's Ligier JS8) just to show the AI where they need to go? If speed isn't relevant when recording.
I've condensed these quotes down to try improve aesthetics but there's a lot wrong with the above:Exactly, speed does not matter, the ai line is not a "ghost lap".
In fact, it is recommended to do it as slowly as possible
To give you a basic idea of how to make an ai line, keep this factor in mind:
The older the car the ai line has to go to the center of the track, modern cars that (above all) brake well, tighter to the curve, for example with a modern f1 you can adjust the curves almost to the maximum, but that line wouldn't fit a car from the 60s-80s. To make an ai line in good conditions you need yes or yes a steering wheel, an ai recorded with a gamepad considers any turn as if you turned the steering wheel completely and this greatly affects the ai, one of the particularities that the assetto corsa ai has to draw curves is the turn of the steering wheel that you apply, the more turn of the steering wheel the slower the curve will do.
1) It's 'technically' true that speed doesn't matter during an AI lap recording. However if you drive slowly then while you'll usually end up with a reliable AI line, it'll be slow as hell. This is simply down to the fact that a racing line at full race speed is different to what you imagine it is when you're driving along slowly. Small subtleties such as how far a car drifts out from the apex when exiting a corner at full throttle aren't captured and you'll find that the AI will be several seconds a lap off their potential pace if you use this method. I've tested this extensively and if you want a fast, competitive AI line then you need to record at maximum speed.
2) The type of car you use is in most cases pretty much an irrelevance. The best driving line for a circuit might well change between a modern F1 car and a 1950s F1 car, but the crux is that you're trying to shorten the distance driven and maximise the speed, factors that apply across every type of car. The key is to record your AI line in a car that you're comfortable driving smoothly at the limit. You're unlikely to create a good AI line using a Maserati 250F, as the nature of such a car means you're constantly making small adjustments to your steering and throttle inputs to keep the car behaving as you want. Something like a modern TCR or GT3 car is a far better option as they're easy to drive smoothly and quickly. Just remember that how different cars drive your recorded line is down to the ai.ini file in the car data folder. It's that file that has a modern F1 car braking at the 50m marker while a 1950s F1 car starts to brake at the 200m marker. Also remember that the default Kunos content is very diverse and all runs fine based on one universal AI line per track.
3) Your actual inputs in recording an AI line are completely irrelevant, it is only the line that matters. The reason you should use a steering wheel if possible is that you'll be smoother and faster. I've just done a test using an Xbox controller to set an AI line using maximum steering inputs where possible. It makes no difference, they just follow the recorded line, none of the AI was at full lock or doing anything stupid.
To summarise:
- Choose a car that you're comfortable driving smooth and fast
- Set a fast, smooth and clean lap
- Make sure the AI sidelines are setup correctly (https://www.racedepartment.com/downloads/ai-line-helper.16016/)
- Add AI hints to any problem areas, examples include bumpy braking areas needing an earlier brake hint, sharp crests needing a maximum speed hint to stop fast cars going airborne, chicanes needing hints to slow cars to avoid accidents etc.
- Make sure the cars you use have correctly setup ai.ini files in their data, use Kunos content as a benchmark.