Something I only noticed recently, though it's quite obvious... raises some questions how to deal with AI offline racing and keeping them within realistic performance to each other. Original ISI content often is "single-make" so it's not visible there (FR3.5, Indy, Meganes...). But when you start to run different model AI cars (like the new ASR 1991 cars or the URD GT cars) it becomes pretty clear that if the basic characteristics of the vehicle are different in real life, and translated truthfully into the game, then the AI will get very different speeds out of them depending on track style and setups. So - hypothetical example - Car A has better grip/better aero, but Car B more straightline speed, and you're looking at a low-downforce circuit, then in real life Team A would need to make a ton of setup adjustments to make up for the design shortcomings of their vehicle whereas Team B could run a pretty standard setup. Contrary on a twisty circuit, Team B has the effort to spend. Now in real life the cars will still compete close with each other because they get their setups adjusted, but most AI for RF2 is delivered with a single default setup. Which means that the gaps between laptimes is much greater than in reality. Of course we can assign setups per team and circuit, but first you have to create them. Which means you end up spending more time as an engineer than as a driver. This would actually work easily if people were willing to share setups for their favorite ride and thus create a pool of "fastest" setups per car per track, but that's not really the case. The other option which is quite a lot easier to handle is like I did for ASR1992 cars yesterday - they all run basically the same physics, even though visually they are modelled like the real-life counterparts - and as such with default AI you have very random results each race, where top teams mix it with minor players and the results look frankly weird. Yet by changing the speed values in the talent file for the drivers, and some other values like Agression and Minskill, you can get them performing pretty accurate compared to each other. You're basically using the AI driver to define car quality. But such cars are boring for the player to drive because there's no distinct personality. It gets even more complicated with "older" Grand Prix because a lot of the driver points/ranking in WDC is based on retirements, i.e. they might have had fast cars sometimes in the top three, but retired more often than a driver with an average car who constantly got a fourth or fifth place. I suppose reliability values are somewhere in the car definitions? Not easy to edit from the outside. For a single race it's not such a big deal, but for an offline championship it would be. This is a criticism or a rant, it's just the observation while RF2 offers many, many "fine tuning screws" to realistically recreate a full season of different cars, it's quite complicated to end up with believable and thus immersive race results versus the AI. Thoughts on this?
Once upon a time I made a complete F1 season for rF1, where I took all this into count. The cars where different,- so I gave them different physics (or setups), all AI-Talent-files carefully adjusted, different "reliability values" as you say, all calibrated to real laptimes on given track from real season! It worked and was thrilling fun! (but a lot of work) A pitty that modders don´t go that deep. We can be happy if they add at least a talent file nowadays! And when a modder does all this (what should be standard) then they where praised into heaven (like CTDP with their 2006 season)!! LOL! But: a waste of time to think about that good old days, as modding (at least for rF2) is somewhere in neverland and surely won´t come up much as long as development stay at this pace. Just my 1000$...
Provided the AI is stable, setups for tracks only need aero & gearing adjustments such that top speed is attained. You don't have to give them the fastest human setup. Thus the engineering effort is not at the same level you might do for yourself. If the cars don't have aero adjustments, then you only need to fuss with the gearing. By all means, the AI talent files should be realistic. Among other things, this stretches out the AI cars so they're not in a big wad. Less of an issue for oval track series, where the cars usually are in close packs, though now you might need two different sets of talent files for series that mix road courses & ovals. Editing the reliability is probably the least desireable choice in my opinion. Partly because engine failure is the only thing modeled and there are a lot of other reasons real cars fail (ball joints, wings, transmission, clutch, fluid leaks, starter motor, etc.).
Regarding stretching out the AI... to my understanding the Speed in AI file only governs the top 10% of AI performance. I.e. a driver with Speed set to 1 will drive at 90% of the general AI strength setting, a driver with speed set to 100 will drive at 100% of the setting. At 100% AI strength, with an AI field of 30, so that each driver is separated from the next by about 3 speed points in the RCD, I had about 6 or 8 drivers within the same second on Spa in F1 1991. That is not a good "stretch"... but the AI file doesn't offer me to go any lower. However, at 85% AI strenght the dispersion becomes realistic, and for those cars that also seems a more achievable laptime for the humans. Good point about Aero and Gearing only for AI relevant parts... might try that. OTOH both the new 1991 Ferrari and McLaren from ASR team reach the same topspeed on Spa at the same brake point, but the Ferrari is still five seconds a lap faster if driven by AI. That's why I assumed there must be mechanical grip involved too, or some other parameters.
What I did for Bathurst with URD & the B****pain Mod: - Created own Talent for all 39 AI Drivers, yes this needed a bit of testing to get a nice Mix and Spread in both mods but it's worth the effort...and I like fiddling! I run them together to have the feeling of "Two Classes" and deliberatley lowered the B****pain mod so all those AI are much slower but still have a nice spread within them. - Adjusted Gearing *only* for all the AI cars, generally just 4th/5th/6th gears. I run the same setup as the AI (URD Aston Martin in this case) as nowadays I only get time for a quick Pick-Up AI Race as opposed to spending hours Practcing and faffing with my own setups! - Made the AIW better as the original was not very good (no before you ask I can't/won't pass this on!) All that for one Track only, when I used to have more time (when I created the Talent Files and Gearing in the first place) I did other Track Specific AI Setups (again Gearing only) but left the AIW well alone as those were ISI tracks where the AIW comes out the box Good! My advice if you have time: 1) Learn the AIW Editor, after a few attempts and learning how to use is properly then a simple FAST PATH and BLOCK PATH takes no more than an hour or so depending on the track size. If you get really good you can do it in less than 20minutes! 2) Create a Talent File if none exists or the original is not good enough, you may need to do a bit of testing and sitting back watching the AI (as above with the AIW) but it is worth it in the long run. 3) See the following thread, it may be "old" but most if not all the information therein (not just on Page 1 but throughout the whole Thread) is still very relevant, worth a read on your lunch break! http://isiforums.net/f/showthread.php/12831-The-definitive-AI-Settings-Thread?highlight=definitive
Don´t read too much from laptimes with a certain mod! There are AI-tyre variables that are most of the time not set correctly by modders! So it often will happen that AI in car A is much faster than in car B, although you (human) do the same times with both cars! Play around more with speed and MinSkill in Talent files. EDIT: AIW has nothing to do with it, as its the same for all AI´s, no matter what car.
Thanks! Good to know. But, who on earth does this???? Like I said in upper post: we can be happy if modders do talent files at all!! For private, if you have the time you can do astonishing things in rF2! Problem: Noone will use it....
