I am wondering if the AI scaled at 100% is unrealistically fast. I have run the FISI/Marussia on Mills at 100%, on both dry and wet, and I'm feeling that the AI is not held to the same driving limitations as we are (on a rail, with very little deviation). I can actually run fairly competitive at 85% strength, and I don't feel like I'm too much of a slouch in the cars. I watch the AI, and it's just incredible, as the strength goes up, they just go into the turns harder, faster, and seemingly stick better than my car can possibly achieve. Don't get me wrong, for a beta, the AI works great, they can race each other and feels like I'm racing people... But at the same time, I'd rather race against drivers, not superheroes. Does anybody feel the same way?
In favor of performance, system wise, probably not. The scalability of the AI makes it usable with a very wide range of skill, or lack off. If you race them at 85 or 90% and have awesome races, definitely stick with that, and very much do not mind that number, it's just what that is. It could very well change per track as well, since it depends on who drove the AI lines as well I can imagine
It certainly does vary track to track. I can beat 105% AI at Sebring, but struggle against 95% AI at Lime Rock. At the end of the day, the % system just allows you to scale their pace, the numbers mean nothing.
Bear in mind here rF2 has a fairly high requirement for the CPU speed; by that I mean if it's not up to speed the game won't be in realtime, and that's largely because of the physics code. Maybe it'll use multiple cores better in future, so there'll be a little more space, but you can imagine having even 5 AI cars using the same physics code is going to be a heck of a drain. On top of that, the more realistic you make the AI physics the better your AI inputs need to be, especially to drive quickly. A simplified AI physics model can take simplified inputs and still look pretty much right. So have x AI cars, with the same physics as the player, and additional processing to create their inputs - not likely at all. (the game would love the player - at least they provide their own inputs) The simplified model is enough, but because there is a fair separation between the player physics and the AI physics - especially the tyres, where only 2 or 3 lines are common - the mod maker needs to do some work to get it close. And then you'd never have all mods feeling like the same difficulty with it set to a certain number, because the AI speed will have been tweaked against different human players for every mod. In other words: what they said Tweak each mod as needed.
Issues with ai are mostly track related. The modders have to create a good aiw file for their track. Just compare ISI tracks to other tracks and you ll see that ai on isi tracks are not too bad.
If modders had tutorials for AIW or at least full spec for all options in AIW editor they would do good AIW. Lot of modders started from rf2 and don't have any experience with rf1. Of course they can try to find some old tutorials for rf1, but lot of those old links are broken. Few months ago Scott promised that he will create full AIW tutorial... Also don't forget that modders don't have as much free time as ISI employers to work on their tracks or cars. That's way they need ISI's help (full specifications or tutorials) .
building GRC track for rF1, my bots didn't jump on big трамлинах, I long selected physics for the bot that wheels would come off a surface, but didn't achieve good result... I was upset that the physics needs to be adjusted for bots, what, at least normally passed turns, on the good, already set trajectory. Also came off a surface on small and big jumps.
The strength differences per track make it very difficult to run a championship (when this game-mode will be available ). To run a championship you might need a AI-strength setting per track, or some sort of slider so you can adjust the AI-speed to your own speed. This would mean you have to test all championship tracks to specify the AI-strength per track. Additionally, I would like to have an adjustable AI balance between qualify- and race-speed, to adjust them to my own speed, so let's say race-strength is 95% of qualify-speed. Last evening I ran a 15-lap race at Malaysia with the Marussia F1s (tire-wear x2). The AI's race-pace was at best 3 seconds slower than their quali-times. The AI's were especially slow at corner 5 and 6 (the long-winding high speed corners), so maybe they were saving their tires, like in real F1. They pitted for tires after 13 out of 15 laps, when their tires were almost destroyed, they were 4 seconds slower than on fresh tires. I wonder if there is some strategy-intelligence so they know they can't finish the race on one set of tires. In that case they would have pitted half way trough the 15-lap race, when the tires weren't 4 seconds slower.
I think this is a big problem with current AI. Even opened a thread about this some time ago to the wishes section.
The issue I found was with the AI strength values found within the .aiw. Specifically the settings for 80%, 100%, and 120% AI strength settings. Essentially, this value is used to scale how fast the AI would be at that circuit at varying AI strength settings. Most rF1 circuits were set at .8, 1.0, 1.2 respectively. It seems that the early ISI circuits for rF2 were set much "stronger" in this setting. Usually .9, 1.1, 1.2. This means your AI will *actually* be at 110% when you set the slider to 100%. This inconsistency between how each circuit is set is *most* of the variance of AI speeds between circuits.
