Development.txt for SHO Competition AI
Jump to navigation
Jump to search
- Development.txt from SHO Competition AI (V14)
- Included here to make the AI tuning details openly accessible on the web
SHO COMPETITION AI
DEVELOPMENT
Version 14 (2022-06-20..2023-06-11)
- DONE: [Location AIW RaceRatio] Imola RaceRatio is tuned too high. I couldn't pass one car in 10 minutes even after setting better-than-qualifying PBs multiple times.
- [README] Removed instructions to set tuned AI Realism (Aggression) value. No need to have a tuned value here anymore. Users can set their own Aggression according to their own preference.
- DONE: [PLR] Reconsider Psychotic AI realism? Not sure it'll help, maybe I'm just too slow for some faster AI. Initial with Timid and Angry vs Psychotic didn't seem to affect qualifying laptimes. Try this after last change...
- DONE: Checked Magny-Cours National with GT cars on Real Aggression. AI were competitive and behaved fine.
- DONE: Checked Imola with GT Cars and Angry Aggression. AI were just as competitive as always.
- DONE: Update README instructions and PLR references to tuned AI Realism value
- DONE: [Location AIW RaceRatios] Check AIWs with RaceRatio < 1.0 (seems odd). Barcelona GP 2003 (FIXED), Oschersleben 2003 (FIXED), Magny-Cours National (FIXED), Anderstorp South (FIXED)
- DONE: Barcelona GP 2003: Seems I used an incorrect RaceRatio that didn't match what was in the tuning spreadsheet.
- DONE: Check Barcelona National: OK. Q: 1:06.886. It's okay now with the proper value from the spreadsheet.
- Was RaceRatio=1.021493813 <-- didn't match spreadsheet RaceRatio=1.1025
- DONE: Oschersleben 2003: OK. PB: 1:23.209. 12th to 8th in 8 laps (10min).
- Was RaceRatio=0.997172531 <-- Didn't match spreadsheet RaceRatio=1.07625
- DONE:: Magny-Cours National: OK. PB: 0.53.540, Q9, 9th until half way (lap 12 or so) when I got lucky with an accident and got to 4th place, that was hard to keep pace with and 5th and behind kept up really close.
- Was RaceRatio=0.997172531 <-- Spreadsheet: 1.07625
- DONE: Anderstorp South: OK. PB: 0:46.553, Q15, 15th to 9th in about 10mins. Could be better but it's okay for now.
- Was RaceRatio=0.990947441 <-- Spreadsheet: 1.06953125
- SCRATCH: Passing too many at Barcelona GP 2003 (33 to 17 after 4 laps). Related to Aggression change from Psychotic to Real? Trying again on Psychotic...no, doesn't seem like it: 9th to 4th in 3 laps on Psychotic.
- SCRATCHED: Doesn't seem to be about Aggression. Fixed with RaceRatio changes above.
- DONE: AI are too strong right off the start as if they don't have to deal with cold tires. Bring back ColdBrain params?
- DONE: RaceColdBrainMin back to RCD original values (instead of all just 1.00). Original values varied between about 0.93 to 0.98, most being 0.94 to 0.96 from what I can see.
- Done: RaceColdBrainTime=45 (previously 0, default 120)
- DONE: Test RaceColdBrainTime=45. May need to change it if it feels off in an actual race.
- Honestly, hard to tell the change here, but I'm happier to be back to or closer to default values.
- [MOD] Renamed MODS\SHO Competition AI\Release Notes.txt to Releases.txt
- [Experiment] Investigated AI Max Load (AI aggression / chance of attempting passes) might be a helpful AI tweak in the future to liven up AI
- Experiment 1: GT's 2004 at Spa 2004
- 0.1 (min): 2:11.911
- 31,500: 2:12.764
^ https://github.com/dbkblk/race07-smartai
- 40,000 (default): 2:12.715
- 1,000,000 (!?): Slower (2:22, +10sec vs 0.1)
- 10,000,000: (2:21)
- 100,000,000: Same
- 1,000,000,000: (2:20)
- 10,000,000,000: (2:21)
- Experiment 2: GT's 2004 at Valencia National
- 0.1 (min): AI1: 1:10.247 (AI100), SB: 1:10.215), Q: 1:10.215, P: 16, ...
- Lot more crashes (mistakes?), a lot more going wide
- 31,500: AI1: 1:10.221 (AI102), AI1: 1:09.839 (AI106), SB: 1:10.100, Q: 1:10.268, P: 21, ...
- Far more normal vs 0.1, far less going wide (none?), less crashes
- https://github.com/dbkblk/race07-smartai
- 40,000: AI1: 1:09.437 (AI100), SB: 1:10.356, Q: 1:10.356, P: 18, ...
- Similar to 31,500
- 80,000: AI1: 1:09.300 (AI110), SB: 1:10.003, Q: 1:10.003, 14, P: ...
- Result:
- Performance: AI Max Load seems like a universal version of AIW RaceRatio. It can increase or decrease an entire field's performance.
- Behaviour: Lower AI Max Load values increase crashes (mistakes?) and going wide while higher values decrease crashes (mistakes?) and going wide
- Updated GTR2 Glossary
- [README] Fixed incorrect Composure value in the manual installation step 2b for Composure. Not a problem if you never used the manual install steps.
Version 13 (2022-06-16..2022-06-20)
- [Testing] Further testing of V13-test2's Composure tuning
- OK: GT 2003: Specials GT's 2003, Chrysler Viper GTS-R, Zuhai GP 2004, AI 100 [SB: 1:30.572, Q: 1:30.572, P: 14, F: 16 (crashed; got up to 8th)]
- OK: GT 2004: Specials GT's 2004, Ferrari 575, Valencia GP 2004, AI 104 [SB: 1:30.032, Q: 1:30.032, P: 14, F: 8]
- OK: NGT 2003: Specials NGT's 2003, Porsche GT3-RS, Imola GP 2004, AI 100 [SB: 1:53.450, Q: 1:53.450, P: 22, F: 16]
- OK: NGT 2004: Specials NGT's 2004, Porsche GT3 RSR, Oschersleben GP 2004, AI 100 [SB: 1:28.404, Q: 1:28.404, P: 6, F: ]
- [MOD] Corrected accidental copy&paste references to "SHO GTR2 XD Style SimHub Overlays"
Version 13-test2 (2022-04-15..2022-06-16)
- [Composure] Increasing Composure in V13-test1 resulted in undesirable retuning of RaceRatio for what seemed like every track. Instead, try a universal decrease of RaceRatio to retain it alreacy custom values from V12. The mistake in V13-test1 was resetting to default RaceRatio and simply adding a static percentage to it thus losing the custom tuning of V12. Instead, this version retains the custom tuning and just decreases the AI performance by tuning AIW RaceRatio with a universal value which will keep the custom tuning of V12 and hopefully prevent having to retune from scratch for each track.
- Problem: Can no longer keep pace with AI who you qualified close to after V13-test1's Composure increase
- Theory: Increased Composure in V13-test1 also increased general AI performance (ie. AI more confident more of the time so they perform better on the whole)
- Experiment: Reduce AI performance by decreasing RaceRatio ideally by a universal amount (ex. -2%)
- Result: OK. Main tests feel decent. I think we're back to V12 quality AI performance with V13-test1's Composure enhancements.
- Testing:
- OK: Specials GT 2004, Lamborghini Murcielago R-GT, Monza GP 2003, AI 96 [Q: 1:45.036, P: 20]
- OK: Specials NGT 2003, Porsche GT3-RS, Valencia National, AI 119 [SB: 1:14.996, Q: 1:15.259, P: 24, F:19]
- OK: Specials NGT 2004, Ferrari 360 Modena, Oschersleben GP 2004, AI 102 [SB:1:29.387 , Q: 1:29.479, P: 18, F: 17]
- OK: Specials GT 2003, Lister Storm, Oschersleben B Course, AI 100 [PB: 0:59.540, SB: 0:59.540, Q: 0:59.540, P: 19, F: 10]
- [License] Added Creative Commons Attribution Non-Commercial License. README updated. License details:
SHO Competition AI © 2021 by Shovas is licensed under Attribution-NonCommercial 4.0 International: http://creativecommons.org/licenses/by-nc/4.0/
SHO Competition AI © 2021 by Shovas is licensed under CC BY-NC 4.0: http://creativecommons.org/licenses/by-nc/4.0/
- [MOD] Split out README Plans section to MODS\SHO Competition AI\Plans.txt
- [Composure Testing] Checking more tracks after V13-test1's Composure + RaceRatio change. This was just testing more tracks using V13-test1's changes - not actually making any new changes here.
- Scratched: I didn't like this approach as it was resulting in effectively retuning every single track again.
- OK: GT 2003: Specials GT's 2003, Chrysler Viper GTS-R, Zuhai GP 2004 [PB: 1:31.000] [Q: 1:31.650]
- Passed 1 lap 2, passed 1 lap 3, passed 1 lap 4, passed 1 lap 5, P20 by lap8/10min, staying with pack ahead, pack behind staying close, couldn't quite pass pack ahead, P23 to P20 by 15min
- OK: GT 2004: Specials GT's 2004, Ferrari 575, Valencia GP 2004 [PB: 1:29.880] [Q: 1:30.795]
- Passed 1 lap 2 (P12), passed 1 lap 3 (P10), crashed back to P27 lap 3, passed 1 lap 3 (P26), passed 2 lap 5 (P24), passed 1 lap 6 (P23), crashed back to P28, quit, pack ahead and behind good
- Observation: Cars far back in the field are very competitive even if you qualified much further ahead then them. Not sure if really realistic or good or bad. When you crash back at least you have some hard racing left to do.
- Theory: They gain Composure from passing you and start doing perfect laps?
- NGT 2003: Specials NGT's 2003, Porsche GT3-RS, Imola GP 2004 [PB: 1:51.690] [Q: 1:53.667]
- OK: Passed 1 lap 3 (P22), caught up to pack (lap 9), could have passed more but crashed lap 9 (P25)
- NGT 2004: Specials NGT's 2004, Porsche GT3 RSR, Oschersleben GP 2004 [PB: 1:29.003]
- Passed 1 lap 3 (P14), crashed lap 5 but was keeping up and could have passed in a few laps
- Other Tests:
- G3 2003: BMW Z3 at Imola GP 2004 [PB: 1:57.826]
- G3 2004: Viper Competition Coupe at Brno GP 2004 [PB: 2:07.114]
- GT3 Cup 2003: Porsche 911 GT3 Cup at Donington GP 2004 [PB: 1:36.027]
- NGT 2004: Specials NGT's 2004, Ferrari 360 Modena, Anderstorp GP 2003 [PB: 1:33.953]
- ...
- NGT 2004: Specials NGT 2004, Porsche GT3-RS, Anderstorp South [Q: 0:49.800]
- NOK: P14 to P4 by lap 8 (5min)
Version 13-test1 (2022-03-08..2022-04-14)
- [Oschersleben GP 2004] Tuning RaceRatio to Qualifying performance as I couldn't keep up
- Problem: AI drive faster when they're ahead of you and slower when they're behind you. This results in AI being harder to catch up to when you're behind them even when you qualified ahead of them. It also results in AI falling farther behind when you pass them than they should even though they qualified ahead of you. Ideas: RCD Composure(?), PLR Player Car Equal(?), PLR AI Max Load(?), ...
- Theory: AI Composure results in AI feeling more confident (faster) when ahead of you versus less confident (slower) when behind you
- Experiment: Try bisection searching Composure
- Reminder: AI Speed Ahead/Behind (Composure) is not the same as keeping up with AI (RaceRatio). AI keeping close to you (Composure) is okay but if you can't keep up to qualifying neighbour positions than you need to tune RaceRatio.
^ I don't think increasing Composure is increasing AI performance (ie. RaceRatio) but AI performance does feel off now.
^ Theory: Higher Composure is making AI more confident more of the time. When they have higher confidence level (when they are ahead of you) they are faster but when they have a lower confidence level (when they are behind you) they are slower.
- Result:
- OK: Specials GT 2004, Lamborghini Murcielago R-GT, Monza GP 2003 [Q: 1:44.912, P: 15]
- DONE: Retuned RaceRatio as I wasn't keeping up to qualifying neighbours
- DONE: +2.5%: Good
- OK: Specials NGT 2003, Porsche GT3-RS, Valencia National [Q: 1:15.079]
- SCRATCH (will try tuning RaceRatio): Note: No RaceRatio tuning here because it's where I tuned Composure
- TODO: +2.5%: Good
- OK: Specials NGT 2004, Ferrari 360 Modena, Oschersleben GP 2004 [Q: 1:29.698]
- DONE: Retuned RaceRatio as I wasn't keeping up to qualifying neighbours
- DONE: +2.5%: Good
- OK: Specials GT 2003, Lister Storm, Oschersleben B Course [Q: 1:00.184]
- DONE: Retuned RaceRatio as I wasn't keeping up to qualifying neighbours
- DONE: +2.5%: Good
Version 12 (2021-12-20..2021-12-20)
- Changes section renamed to Release Notes and moved to MODS\SHO Competition AI\Release Notes.txt
- Development section moved to MODS\SHO Competition AI\Development.txt
- Problem: (Continued) Cars are off qualifying pace by too much in race sessions. Results in passing too many who were faster than you in qualifying.
- Theory: Track AIW RaceRatio is not representative of AI's qualifying performance
- Experiment: Adjust Track AIW RaceRatio to better represent qualifying performance
- Result:
- OK: Specials GT 2004, Lister Storm, Spa GP 2004 [Q: 2:16.763]
- OK: Specials NGT 2003, Porsche GT3-RS, Valencia National [Q: 1:15.602]
- OK: Specials NGT 2004, Porsche GT3-RS, Valencia Long [Q: 1:55.763]
Version 11 (2021-12-18..2021-12-19)
- Problem: (Continued) Cars are off qualifying pace by too much in race sessions. Results in passing too many who were faster than you in qualifying.
- Theory: Track AIW RaceRatio is not representative of AI's qualifying performance
- Experiment: Adjust Track AIW RaceRatio to better represent qualifying performance
- Result:
- OK: Specials GT 2004, Lamborghini Murcielago R-GT, Monza GP 2003 [Q: 1:45.072]
- OK: Specials NGT 2003, Ferrari 360 Modena, Monza Junior [Q: 0:55.124]
- OK: Specials NGT 2004, Ferrari 360 Modena, Oschersleben GP 2004 [Q: 1:29.799]
- OK: Specials GT 2003, Lister Storm, Oschersleben B Course [Q: 0:59.295]
Version 10 (2021-12-16..2021-12-17)
- Problem: (Continued) Cars are off qualifying pace by too much in race sessions. Results in passing too many who were faster than you in qualifying.
- Theory: Track AIW RaceRatio is not representative of AI's qualifying performance
- Experiment: Adjust Track AIW RaceRatio to better represent qualifying performance
- Result:
- OK: Specials GT 2003, Ferrari 575 GTC, Dubai Club [Q: 0:59.034]
- OK: Specials GT 2004, Ferrari 575 GTC, Dubai International [Q: 1:32.053]
- OK: Specials NGT 2003, Porsche GT3-RS, Dubai National [Q: 1:25.707]
- NOK (AI Broken): Hockenheim National
- NOK (AI Broken): Hockenheim Short
- OK: Specials NGT 2004, Nissan 370Z, Magny-Cours GP 2003 [Q: 1:42.790]
- OK: Specials GT 2003, Lamborghini Murcielago R-GT, Magny-Cours National [Q: 0:53.381]
Version 9 (2021-12-09..2021-12-16)
- Problem: (Continued) Cars are off qualifying pace by too much in race sessions. Results in passing too many who were faster than you in qualifying.
- Theory: Track AIW RaceRatio is not representative of AI's qualifying performance
- Experiment: Adjust Track AIW RaceRatio to better represent qualifying performance
- Result:
- OK: Specials GT 2003, Chrysler Viper GTS-R, Anderstorp South [Q: 46.977]
- OK: Specials GT 2004, Chrysler Viper GTS-R, Barcelona National [Q: 1:07.780]
- OK: Specials NGT 2003, Ferrari 360 Modena, Brno GP 2003 [Q: 2:04.554]
- OK: Specials NGT 2004, Ferrari 360 Modena, Donington GP 2004 [Q: 1:33.949]
- OK: Specials GT 2003, Ferrari 550 Maranello, Donington National [Q: 1:04.227]
Version 8 (2021-12-01..2021-12-09)
- Problem: (Continued) Cars are off qualifying pace by too much in race sessions. Results in passing too many who were faster than you in qualifying.
- Theory: Track AIW RaceRatio is not representative of AI's qualifying performance
- Experiment: Adjust Track AIW RaceRatio to better represent qualifying performance
- Result:
- OK: Specials GT's 2003, Porsche 911 GT2, Dubai GP 2004
- OK: Specials GT's 2004, Maserati MCI2, Spa GP 2003
- OK: Specials NGT 2003, Porsche GT3 RS, Enna GP 2003 [Q: 1:42.085]
- OK: Specials GT's 2003, Saleen S7-R, Oschersleben GP 2003 [Q: 1:23.554]
- OK: Specials NGT 2004, Ferrari 360 GTC, Estoril GP 2003 [Q: 1:41.992]
- OK: Specials GT 2004, Chrysler Viper GTS-R, Monza GP 2004 [Q: 1:45.828]
- OK: Specials NGT 2003, Porsche GT3-RS, Imola GP 2004 [Q: 1:53.128]
- OK: Specials NGT 2004, Porsche GT3-RS, Magny-Cours GP 2004 [Q: 1:41.862]
Version 7 (2021-11-29..2021-11-30)
- Problem: (Continued) Cars are off qualifying pace by too much in race sessions. Results in passing too many who were faster than you in qualifying.
- Theory: Track AIW RaceRatio is not representative of AI's qualifying performance
- Experiment: Adjust Track AIW RaceRatio to better represent qualifying performance
- Result:
- OK: Specials GT's 2003, Lister Storm, Barcelona GP 2003
- NOK: Specials GT's 2004, Lamborghini Murcielago R-GT, Hockenheim GP 2004: Broken; Cars go around at pit speed; Even happens on my vanilla install; Not sure what's up;
- OK: Specials GT's 2004, Lamborghini Murcielago R-GT, Zuhai GP 2004
- OK: Specials NGT 2004, Nissan 350Z, Brno GP 2003
- Investigate: Do a proper bisection Search on AI Max Load to find the behaviour and the optimum point as I still think the AI get better of AI-Player collisions (ie. it effects them less than you)
- Experiment: Bi-section search AI Max Load
- Result: NOK: I had this wrong. AI Max Load isn't about collisions. It seems to have to do with AI aggression and chance of attempting passes. Might be interesting but for later.
- Experiment: Investigate Car HDC "AI Torque Stab" as I think this has to do with weight
- Result: NOK: Less than 1.0 had problems while 1.0 acted basically realistically but not sure I could tell a different between 1.0 and 1.25 (the default). So to preserve progress with AI performance tuning in races I'll just let this be for now and return to it if the need arises.
Version 6 (2021-11-25..2021-11-29)
- Problem: Cars are off qualifying pace by too much in race sessions. Results in passing too many who were faster than you in qualifying.
- Theory: Track AIW RaceRatio is not representative of AI's qualifying performance
- Experiment: Adjust Track AIW RaceRatio to better represent qualifying performance
- Result: OK - Not perfect but much, much better than default.
- Specials NGT's 2004, Porsche GT3 RSR, Anderstorp GP 2003: OK
- Specials NGT's 2003, Ferrari 360 Modena, Valencia GP 2004: OK
- Specials GT's 2003, Chrysler Viper GTS-R, Donington GP 2003: OK
Version 5 (2021-11-25..2021-11-25)
- Investgate: "AI to AI Collision Rate" (Default 20, V1 40) might be impacting AI in some way. Like perhaps 40 is too complex and results in wonky AI decisions. So try a bisection search to understand how it's effecting AI.
- Experiment: Bi-section search and note observations
- Result: NOK: It's not obvious going above the default (20) actually helps. So leave it at 20 for now.
- 1 (Min): Oddly not as bad as you would think. Yes they touch a little bit more on starts and first corner but after that it's a touch once in a while not the wreck fest you would expect.
- 40: This is my usual. They seem fine but you still get those start and T1 collisions. Not sure if other values would improve upon this but it's worth a try.
- 20: Fine, I guess. A number of touches on start and T1 but nothing crazy.
- 10: Definitely more touches on start and T1. So I guess more is better.
- 30: Less touches than 10 definitely
- 40: Weird. Plenty of touches with GT's at Valencia GP 2004, messy T1, sometimes full or near stoppages at T2. Maybe there's something up with too much of this param.
- 20: Definitely better (not saying it's perfect but better) on start and through T1
- 30: Pile up T1 the first time, second and third starts definitely slowdowns T1 and T2
- Problem: V3 Pyschotic AI Realism=2.0 results in AI ramming you from behind if you're too slow in a braking zone or corner for example but it's not really realistice
- Theory: AI Realism is too high causing AI to act more aggressive than is realistic
- Experiment: Lower AI Realism to lower AI aggression which should lower the frequency of ramming due to being slower than AI
- Result: OK: Seems like an improvement in sportsmanship
- OK: Specials GT's 2004, Ferrari 575, Valencia GP 2004
- 0.7 (Timid); Specials GT's 2004, Ferrari 575, Valencia GP 2004; Less abrupt maneuvers in straight lines where they might have side contact; Seem to run wide when you're on the inside to avoid hitting you; Maybe(?) not using full throttle (I'm overtaking cars on wide open straights I don't think I should be); They do still pass, though, when they have a clear opening, but they don't force tight passes;
- 1.35 (little bit more than Angry=1.30): This is called "Psychotic" in-game but it's very close to Angry (1.3); AI are behaving; They don't give up spots too easily; No big contacts yet; Best AI fell behind 6sec+ by mid-race but this is probably due to something else (eg. Track AIW QualRatio vs RaceRatio - problem logged)
- 1.675: Qualified first but decided to run second first lap then discovered the leader was very hard to pass in a race-worthy fashion (a good thing); Third really kept up nicely within 0.5sec; No real reckless driving to speak of; Positive experience overall, would be nice to get even more competitive with them;
- 1.8375: Passed a few but they became difficult after a while; A good improvement I think; I think I'll move on as this feels like a point of diminishing returns.
- OK: Specials NGT's 2004, Porsche GT3 RSR, Oschersleben GP 2004
Version 4 (2021-11-21..2021-11-24)
- Problem: AI fall too far behind when passed / AI never catch up after passing big pileups with AI that are definitely better than you (same root cause?) but AI that you didn't pass keep ahead of you and very competitive
- Theory: Talent RCD Composure levels too low result in loss of composure after pass and they fall behind
- Experiment: Increase Talent RCD Composure levels to reduce loss of composure after pass so they keep up
- Result: NOK: Hard to tell if there's a difference. Composure effect is very subtle or not doing anything at all.
- Test 1 (Composure=1.0): Specials NGT's 2004, Porsche GT3-RS, Imola GP 2004 / Passing cars fell behind by 1-2 seconds but seemed to be able to keep it to just 1-2 seconds / Oddly smooth T1/L1 but did notice minor errors before T1 / Boring AI?
- Test 2 (Composure=0.5): ~ / Passed a pileup, I guess they caught up
- Test 3 (Composure=0.25): ~ / Hard to say...
- Test 4 (Composure=0.001): ~ / Hard to say again...
- Test 5 (Composure=1.0, again): ~ / Just so difficult to tell if there's any change...
- Theory: Talent RCD MinRacingSkill default is too low allowing drivers to decrease in skill by an unrealistic amount (ie. 10%)
- Experiment: Increase Talent RCD MinRacingSkill to see if skill level is causing the issue when they fall behind
- Result: NOK: Once I understood MinRacingSkill + Composure and how they worked together I started tuning Composure instead (see next Theory).
- 1.0: Definitely seemed like they were 100% skill the entire time / not sure if that's absolute or relative to their own RCD parameters like Composure or RaceAbility etc
- 1.0 + Composure=1.0: Felt like they were losing composure more but they kept up after 10sec or so
- 0.0 + Composure=1.0: lol 2-3 incidents before T1 and 1 immediately after T1 / So yeah MinRacingSkill=0.0 and Composure=1.0 means basically no skill at all at times
- 0.0 + Composure=0.0: Okay, so Composure=0.0 seems to be perfect composure / Even though MinRacingSkill=0 (bad) the AI never felt like they lost their skill / So Composure = 0-100 it seems and MinRacingSkill = 0.0-1.0
- 0.90 (Default) + Composure=0.005: ...
- Theory: Talent RCD Composure default is too high (0=Perfect) causing AI to make too many mistakes and cost them time
- Experiment: Lower Talent RCD Composure so that AI make less mistakes and cost less time
- Result: OK: I think this is a decent starting point. Can come back to this later after tuning other areas.
- OK: Specials GT's 2004, Ferrari 550, Monza GP 2004; Past a few too many peopl for my liking, 11 or so, but at some point the AI got good enough and it became more and more difficult to make up spots. I think AI who had fallen far behind started catching up again after a while. If other results are OK then maybe this will have to be a starting point for a release now and improvements later.
^ Note: Ferrari 550 seems overpowered. It was overtaking everybody on straights and seemed faster through corners.
- OK: Specials GT's 2004, Ferrari 575, Valencia GP 2004; Passed 5 within a few laps and then the AI really started to put up a fight; The guy behind kept to within 1.0-1.2sec mostly.
- OK: Specials NGT's 2003, Porsche GT3-RS, Imola GP 2004; Intense race actually. Again, made up 7 spots or so starting from 17th is a little too easy than I would like but what can you.
Version 3 (2021-11-21..2021-11-21)
- Problem: AI too nice, too easy to intimidate them to let you pass (
- Theory: AI Realism too low (Default 1.0 "Real") resulting in "timid" Aim
- Theory: Talent RCD related? Aggression or Composure?
- Experiment: Bisection search AI Realism (Max 2.0)
- Result: OK; Seems like a positive change. Might have to review later if I notice too many crashes or pileups, though.
- OK: Specials GT's 2004, Ferraro 575, Valencia GP 2004
- OK: Specials NGT's 2004, Porsche GT3-RS, Imola GP 2004 [Q: 1:53.605]
- One big pile up, seems I heard Crew Chief mention more crashes than I would expect. Might be a side effect of "Psychotic" AI
- Note: This might be causing more crashes and pileups. Might have to back it off if I find it's too much of a problem.
- Problem: Still seeing slow cornering in some scenarios / tracks. Eg. GTs 2003 at Anderstorp GP 2003. Note I'm not sure I saw problematic slow cornering in other test cases. It might be Anderstorp is an outlier and requires a more specific track fix (eg. AIW lines).
- Theory: AI Corner Grip causes AI to think they don't have enough grip on corners to go faster
- Experiment: Increase AI Corner Grip to allow AI to use more grip which should allow faster cornering
- Result: Cornering speeds seem fine at tested tracks.
- OK: Specials GT 2004, Lister Storm, Zuhai GP 2004
- OK: Specials NGT's 2004, Porsche GT3-RS, Imola GP 2004
Version 2 (2021-11-20..2021-11-21)
- Problem: AI slow off green, slow first lap, etc. (ie. ColdBrain params)
- Theory: ColdBrain effect resulting in sub-optimal AI behaviour
- Experiment: Disable ColdBrain effect in Talent RCD files
- s/RaceColdBrainMin\s*=\s*(-?[\d\.]+)/RaceColdBrainMin=1.00 // DEFAULT=\1 /
- s/RaceColdBrainTime\s*=\s*(-?[\d\.]+)/RaceColdBrainTime=0 // DEFAULT=\1 /
- Result: OK; AI seem to be full speed immediately which is what I wanted.
- Note, only for Race session. ColdBrain still at default for Qualifying which is fine since you can time accelerate there. Might want to consider it for Qualifying later though just so you don't even have to do that.
- Problem: AI still sometimes braking early but also sometimes seems good (GTs 2003 at Anderstorp GP 2003)
- Theory: AI Brake Power and Brake Grip values causing AI to think they have to start braking earlier
- Experiment: Increase AI Brake Power and Brake Grip to allow AI to stop later
- Result: OK; AI braking points seem okay. Might be a little late to be realistic but it makes them competitive. Might reconsider later.
Version 1 (2021-11-19..2021-11-20)
- Tuning UserData PLR AI Brake Power and AI Brake Grip params for alleviating early braking by AI
- Tuning UserData PLR AI Corner Grip param to alleviate slow corner speed by AI