Separate names with a comma.
Discussion in 'rFactor 2' started by Kevin Knorpp, Dec 28, 2016.
Please share combos and settings (aggression, etc.) you are using for great offline races.
Well here's one with the amazing Flat 6 at Palm Springs Road Course A. I ran a race with Empty's settings...
... and had huge fun
I'll upload a video to RaceTube and post a link here to my results after a few more practice races
Great idea Kevin, but the ai strength settings will show more accuracy with only you or some one who has a similar skill level as you doing the tests because of the varing degree of the human skill levels.
Note: If you need to setup the ai strength to more than 100% then this is usually a good indication that either or both the aiw and the car mod's ai settings need attention.
Interesting...your two points above seem to be contradictory?
Personally, I don't think AI will ever be as fast as drivers at least in my lifetime (not long lol) and this is doubly true if any decent driver actually concentrates on a track even a little bit even with practice only (i.e. not setup testing or data analysis) and even with default setup and even default fuel.
Maybe I should say that differently... I have never had expectations that AI would ever keep up with me with any setup testing beyond simply balancing the vehicle and certainly not data analysis, etc. which I never do.
What I think is reasonable is for the AI to be able to challenge the average mid pack league racer for at least long enough to practice accident avoidance, awareness, etc. for a league race, for example.
In my experience AI is rarely good enough at 100% even for the beginning stages of this let alone any serious time in any combo which is why I have to rotate content so much offline but that is OK and also why leagues are so important for people like me who want to focus on a single vehicle and a single combo for any amount of time.
Properly calibrated AI at 100% will be very difficult to beat unless you are both extremely fast and extremely consistent. Let's say the top 5% of sim racers fall into that skill group. Many of us cannot do it on a regular basis, but have the occasional lucky day or session. Most cannot do it at all, ever.
You should expect to use AI at less than 100% until you reach that 95th percentile status.
The problem is, as you well know given your posts here, that the AI are not calibrated to be relative, nor are they even programmed properly at all at many tracks. So it is a complete crap shoot as to what level you need to set them at to be relative to your own skill. I can beat AI at 100% by some fraction and be first place fairly easily at some tracks. Others, I shouldn't even be in the same race with the AI at 100% and am just getting in their way. My general skill level is the same in both scenarios and I am using the same car, so after 100 laps at a track if I am way out ahead or way behind a cluster of 20 AI that themselves show some range of speeds, it's the programming, not me. Don't even get started when you use different cars at the same track and get the same wildly divergent results.
Give-up now. Use the slider to make for competitive racing--for you, in the car you choose at the track you choose. It will be different for every combination and you will improve over time as well. You still have to take this approach in AMS and they actually try to calibrate the AI. There has been no attempt here with rF2 to date.
I been racing online and offline plus improving the AI fast lines (AIW) since day one in rF1 (2005) and now doing the same in rF2. I can assure you that from my race experiences expanding from the early days of SCGT (1999) that after i make the best AI fast line (AIW) that is humanly possible and fine-tune the car mod's AI settings (if necessary) plus assign the AI a car setup then i will have the race of my life and usually have no chance in hell of winning with the AI strength set to 100%.
In my AI races i will always race them fairly with no pushing them out of my way to pass or any risky dive bombing. I race them as i would against all human drivers and the way i would want to be treated. With this approach i get so engrossed in my race that i sometimes forget that i'am actually competing against an AI.
I think we are actually agreeing because I'm saying that I don't expect 100% to be fast enough and you guys are saying 100% should theoretically be fast enough but never is without specific tuning of everything for a specific combo.
No, what we're saying is 100% AI, if properly calibrated, is too fast because we are not Michael Schumacher. The AI are "perfect" unless specifically programmed through talent settings to make mistakes.
We can do a perfect lap once in a while. They can do them consistently. That's why most people prefer racing against humans. The rF2 AI are less robotic and more aware than other AI, but at 100% we are saying we want a perfect (100% on the test score) opponent.
Are you perfect, or an A student or a B student or maybe a beginner C? Not many "perfects" out there in the real world, even in real racing, but in a sim it is useful to have a 100% setting for extreme test purposes.
Unfortunately, due to lack of calibration, you might have to set the AI at 110% or 90% to achieve Michael Schumacher skill levels...you never know.
Sorry I'm just confused now because the issue at hand is whether it is useful to share offline race settings (including AI%) which I believe is useful as a mid-pack racer reference.
Let's assume I tell you that at track A, I think Clio AI set at 93% is equal to my skill level and that the AI are a bit too aggressive. Out of the thousands of combinations available, let's say another person mentions their suggestion is 101% for the same track and car combo, but that person says the AI behave themselves nicely.
What is the baseline? My speed level, or the second person who is obviously faster than me? Are the AI too aggressive because I set them above the default, or because I am obviously too slow at that track, or both, or for other reasons?
Go ahead and start this database if you want, but there is no consistent baseline (driver skill; AI %, AI Aggression %, number of laps spent practising, real road levels, weather, etc.) from which to draw any useful conclusions. And that's if you can get people to contribute the same car and track combinations, of which there are thousands.
S397 could completely reprogramm a method to incorporate the AI Learning into a track and car's "memory" before you will have even 5% of the info you think you need to be useful.
But it's OK if you don't want to accept that the AI slider needs to be arbitrarily tweaked for every car and track combo. At least we have that option in the UI. For the first eight or nine years of rF1 and rF2 we didn't even have that.
Marc.....this is so true, great explanation for the layman!
When I poked at creating a similar database for myself, I mainly noticed that there are a few tracks where the AI is seriously slow no matter what car is chosen. The other factor was whether creating an AI gearing/aero config was needed, assigning a setup to the AI.
It is relatively easy to make any of the cars have impossible-to-defeat AI by increasing AIGripMult in the TBC and then repackaging the mod. The side effect is that when you go much higher than 1.10, then you'll notice how superglued the AI is when you bump them.
You guys are overthinking this. Just share some settings. It's not a database and I won't hunt you down and kill you if it's not the best race I ever had.