53 replies to this topic

[jgf]

This confusing array of 60fps patches, with one working for some mods but not others, one may have functioning AI in these mods but not those, etc., is what kept me from using them, except in testing mode.

FWIW, from my understanding of the LP files, they are based on one record for each "tick" (36 per second ...thus the game's 36fps default limit?) longitudinally around the track;  I assume allowing 60fps has the game running at 60 "ticks" thus confusing the AI.  Would creating new LP files based on 60 ticks alleviate this problem?  Or would it not work at all?  If it did work, it would probably entail laborious manual editing of at least each race.lp.

I've been running the '65 cars lately and, with the newmod patch enabled, thought it odd that in practice the AI run normal speeds and only slow in the race itself - at Laguna I'm 8th in practice, about 8 seconds behind the leader, but will lap the field in a 12 lap race;  disable the patch and the AI run similar times in practice and race.

[Saiph]

M Needforspeed, on May 05 2014 - 03:04 PM, said:

Real life and I can 't read all the interesting posts above, but you may want to have a look at this
https://srmz.racesimcentral.net/inde...288 and here https://srmz.racesimcentral.net/inde...t=15#entry55014
Whatever your target,of all the patches version released,  60fpsaiv1 seems the way to go for.
You can tweak the AI files for all mods, and you don't get the ugly look of unrealistic ride heights

That's interesting. I couldn't remember hearing about 60fpsaiv1 before, I just checked back in the patch thread to find Petteri's comment about it. But as I've said before, my main experience is with solo testing, so if you've found that it's better for racing against the AI then I bow to your greater experience in that area.

[Alan Davies]

I have not used the 60FPS patches before as I race mostly against he ai; but having seen in the thread about the 60fpsaiv1 I gave it a try.
Although the frame rates do not improve (using Fraps - about 30 ish) it does seem smoother though.  The ai are slow and trying to speed them up become VERY aggressive.

[dbell84]

jgf, on May 05 2014 - 03:16 PM, said:

This confusing array of 60fps patches, with one working for some mods but not others, one may have functioning AI in these mods but not those, etc., is what kept me from using them, except in testing mode.

FWIW, from my understanding of the LP files, they are based on one record for each "tick" (36 per second ...thus the game's 36fps default limit?) longitudinally around the track;  I assume allowing 60fps has the game running at 60 "ticks" thus confusing the AI.  Would creating new LP files based on 60 ticks alleviate this problem?  Or would it not work at all?  If it did work, it would probably entail laborious manual editing of at least each race.lp.

I've been running the '65 cars lately and, with the newmod patch enabled, thought it odd that in practice the AI run normal speeds and only slow in the race itself - at Laguna I'm 8th in practice, about 8 seconds behind the leader, but will lap the field in a 12 lap race;  disable the patch and the AI run similar times in practice and race.

60fpsaiv1 will work with 67's or any mod, so remembering which one of the others works with what mod isn't necessary.  The only thing I'm not sure of with 60fpsaiv1 is if it works with online racing.  I think one of the others is needed for that.

[jgf]

I just tried the 60fpsaiv1 patch.  The AI were decidedly slower, changing no other settings I went from 8 seconds down in practice to leading by 3 seconds.  Speeding them up I tried again, was 5 seconds down in practice but still lapped the field in a 12 lap race.

It still seems to me the major effect of the 60fps patches is on the AI race performance, not their practice times.  But aren't both controlled by the same parameters?

[Roger_F]

Fellow enthusiasts, I know this is an old thread but the topic is still very interesting.  60fps would likely be the design target if GPL were developed during the last 10 years rather than in 1998 or so.  I think it is generally accepted that gpl looks much better at 60fps than at 36, so the focus on how to get this to work still seems worth the effort.

I don't have the gpl / programming design skills to build another 60 fps patch so my approach has been to read everything I can find on the subject and start experimenting from there.  I don't race online so my focus has been on racing against the AI.  Others of you have described the 60fps AI problem and a reasonable summary is probably as follows:  (1) the AI are too quick off the line at the start of the race; (2) the AI are too fast in slow corners; (3) the AI are too slow on long fast straights; and (4) last but not least  ... they are way  too aggressive almost everywhere if you try to speed them up to normal lap times.

Several great suggestions have been put forth that solve some of the problems: (a) use the 60fps ai1v patch (it adjusts the track parameters to try and adjust for the frame rate difference and seems to run on all mods); ( reset the AI parameters in GEM 2 to about .85 to speed up the AI average lap speed to something close to the 36fps levels and live with the rest of the over under problems (i.e. start from the back of the pack and enjoy passing a good portion of the field during a short to medium level race).

I have experimented with a few other adjustments that seem to improve the situation substantially.  First off use the 78 patch in GEM 2 that slows the AI in the corners by 10% (78l Ai Speed Control V2) ... this could even be 15 or 20% for our purposes.  Secondly, go to the driver.ini for the installation that you have (e.g. one overall installation; individual installation per mod; or race specific for the 55 MOD or the 67 Historical Mod and adjust the aggressiveness index for each driver in the field to approximately 80% of its current value.  These values are typically, aggression = 1.030000 or less and change them to 0.824000 or less.  Adjust every driver in the potential field (usually 19 in total).  The net result is substantially better than before but still not perfect.  Hunt for driver,ini or driver55,ini or driver65.ini etc.; they all open with notepad.  You should make a copy of each driver.ini before you start experimenting ... just to be safe.

Typically the tracks with the most extremes (fast straights leading into very slow corners) have the most problems.  These two additional corrections reduce the number of crashes caused by the AIs and allow for a reasonable race with much lower probability of you being "taken out" by the aggressive AI before the race is over.  Rouen is a great track to test on due to its very fast straights leading into very slow corners.  I'm still experimenting ... maybe more to follow!

At 80% the AI are still pretty agressive but not ridiculas; ... there is a happy medium that is a best compromise for each individual track but one overall setting across all of the many gpl tracks is a much greater challenge.  Greg's race specific car sets for the 55 mod is an ideal solution space, because you can adjust each track separately ( i.e. each track has its own driver55.ini file); unfortunately not every Mod has this feature.

Thanks for your time, hope this helps some of you to better enjoy racing the AI in gpl at 60fps.  As in real racing, the further down the track you can focus, the faster you can drive.

Roger

[Roger_F]

A follow up on my last post (#21) re: 60fps AI racing.  The scenario described above applies mostly to F! cars and tracks from the 55,65,67, and 69 Mods.  To better understand the overall challenge of using the 60fps ai1v patch I ventured away from F1 and towards CAN AM (71 and 66 Mods).  I started by setting the GEM 2 AI at .85 (this speeds up the AI by maybe 15 to 20%) and turning on the 78l Ai Speed Control V2 (this has the effect of slowing the AI down a little).  The result was different than I expected!

The 71 CAN AM cars were too slow in the corners but OK on the straights ( I turned off the AI speed control patch in GEM 2)?   Mid Ohio CA71 was a little slow as was Road Atlanta 71 at the .85 GEM 2 AI setting.  I adjusted the dlong_speed_adj_coeff in the respective CAtrack.ini files up about 5% for those two tracks and enjoyed the racing.  The 10 championship courses for the 71 Mod are generally more wide open and have less severe corners than most of the F1 courses (Mosport being an obvious exception): that may be part of the explanation for the reduced incidence of AI crashes.

The 66 CAN AM cars ran best with the GEM 2 AI set at 1.00 and the 78l Ai Speed Control V2 turned off.   In other words, just using the 60fps ai1v patch as is?  That includes using the special gpl_ai,ini supplied with the patch (on all Mods).  I tested on the 6 championship tracks recommended in the 66 MOD plus Nassau.  I read elsewhere that this mod needs to be run with the Gem 2 AI set at 1.0.  I'm not sure why this still works with the 60fps patch but that seems to be the case?

Both CAN AM fields are set with a wide performance difference slowest to fastest drivers & cars and this may have contributed to a reduction in AI crashes?  The wide open tracks may have also been a factor ... I decided not to adjust the driver aggression stats.  

I'm not sure I fully understand these results but the implication seems to be ... that all Mods are not created equal?  The race starts (at 60fps) are faster than I'd like on both CAN AM Mods, but other than that you can have a pretty good race against the CAN AM AI with the above suggested settings.

My testing objective is to have the AI field approximately half above and half below a comfortable personal qualifying speed.  Also factor in to your understanding that I'm probably an average GPL driver.  If you are faster or slower you may have to adjust the GEM 2 AI factor or the dlong_speed_adj_coeff, appropriately.  Instead of adjusting the GEM 2 AI you can also directly adjust the NPT_override setting found in the 60fps modified gpl_ai.ini file.  On the base 1967 program the GEM 2 adjustment does not seem to work, so you have to resort to a direct NPT adjustment..

I hope this information helps; if you get different results let me know.  I am still pursuing the 60fps quest ... i'll post again if I come up with anything else that may be useful.

Roger

[mysterion]

Hi Roger,

I just wanted to thank you for your work on this issue which has really plagued my offline experience of GPL at 60fps (which I agree is the only real acceptable baseline now).

Strangely I never really had any 60fps AI issues until I installed the Gpl 67 Patch V1.3  - I actually thought I was improving at Spa until I realised the AI were just dog slow!

Please do post any further findings you discover.

- David

[Roger_F]

David,

Thanks for the encouragement!

Good solutions seem to be MOD specific and in some cases track specific within Mods.  Increasing overall track AI performance seems to bring with it increased driver aggression.  On some tracks it is severe enough that you have to drive extremely defensively to avoid being "taken out" by the AI.  Controlling driver aggression (at those tracks) seems to be one of the larger challenges.

Roger

[JonnyA]

You can mitigate against very aggresive AI by raising their Awareness values. It makes them less likely to crash into you.

[Roger_F]

JonnyA,

Good to know!

Any idea how sensitive the attribute is?  Setting the AI at .85 (in GEM 2)  seems to be a common 60fps adjustment (1.00 / 0.85 = 17.6% increase in AI adjustment value); does alertness need a similar percentage adjustment to reduce crashes?  Also what else does this attribute effect?

Thanks,

Roger

[JonnyA]

Try awareness values around 1.2

[Roger_F]

This represents the next step in the continuing process of converting my GPL Mod collection to 60fps (a follow up on posts #21 and #22).  I decided to tackle the 1967 Historical Mod in all its magnificent splendor.    This very large collection of different cars / drivers / tracks all need to be addressed in some way to make 60fps work.

I decided a game plan was in order:

(1) Take one or two tracks I know rather well and see how the AI react to the 60fps ai v1 patch combined with the 78l Ai Speed Control V2 Patch.  This combination of patches seems to work best on F1 cars and tracks and should serve as a baseline for development ... at least this was my hope.  

(2) From this mini test formulate a larger approach (if it still made sense) to the 18 or so tracks included in the mod; This assortment of tracks has 7 different sets of cars available based on the time of year the race was held.  It also has three different sets of driver lineups.

(3) Test for an overall Gem 2 AI setting for all 18 tracks ... if one exists.

(4) If one general Ai setting is close (e.g..85, or .90, or .95, or 1.00 etc.), then test (through AI simulation qualifying sessions) for needed  adjustments to the dlong_speed_adj_coeff in the respective 18 track67.ini files.  I chose to target Ai pole position values at 10% above published Rank Historic hot lap times.  These times are generally close to the actual pole position times recorded for the races we are discussing.  For the non championship races I used the actual pole position times plus 10%, as I could not find the Historic Rank times for those tracks.

(5) if adjustments to driver.ini files are necessary; figure out how to handle the 3 sets of driver lineups included in the Mod ... I was hoping against all hope that this step was not necessary as these tracks share driver lineups and getting a good compromise between tracks might prove to be very difficult.

Long story short, step (5) was not necessary!!!!!!

I probably ran 90 or so simulation qualifying sessions using novice race specs and advancing the 15 minute qualifying time through the advance box tab.  It only takes a few minutes, under this process, to determine an AI qualifying field.  Again my target was to get the top AI qualifyer to be about 10% above the Rank Historic Time for the track.  Each track was adjusted individually to operate under one overall Gem 2 AI value.  I chose 1.00 (after much additional testing) as my Gem 2 AI adjustment value; individual adjustments to the track67.ini dlong_speed_adj_coeff (s) were generally within 5% of the original values.  This process generally puts me mid field during qualifying.  If you are faster or slower you will have to adjust accordingly.  For instance, if you can set repeatable Rank Historic qualifying times, GEM 2 AI would be set to approximately .90 and you would vie for pole position in every race.

This process then allows one overall future adjustment of the Gem 2 AI value to effect all 18 tracks equally.  This whole simulation / adjustment process took about 8 hours of work but the results were well worth the effort.  As I improve, I can adjust the whole 67 Historic Mod through one Gem 2 AI value ... in just seconds.  GEM 2 remembers the AI value from race to race, because it is attached to the mod unique driver name.  I basically race at 60fps just like I did at 36fps; I'm also able to increase both detail level and screen resolution without any additional stability issues.

This all sounds more complicated than it really was to execute.  The hard part was figuring out the process.  Knowing what I know now, I could repeat this whole process in less than half the time.

The saving grace was that I did not have to adjust the aggression or awareness indices of each driver.  Tampering with these indices would have increased the effort level substantially.

I again, hope this helps those of you that are considering a similar journey.  I'm slowly converting my 12 GPL Mods to all be 60fps AI capable.

Roger

[pioujd428]

This looks interesting Roger. When I get a chance I will try and follow your steps to see if I can get the ai to work reasonably for myself at 60fps.

Maybe you could post your track67.inis and other files you have edited? I think I would like their speed the same as what you have set.

[Roger_F]

pioujd428,

Here are the pertinent data:
<
Track /    car set number   / dlong_speed_adj_coeff (in track67.ini)   /   target AI pole position
Kyalami67/ 1 /    1.06400  /    91 sec
Brads /    1 /    1.03000  /   102 sec   
Oulton -Spring  / 1 /    1.01000  / 101 sec
Silver2Express / 1 /    1.01000  /   100 sec
Moncane / 2 /    1.00410  /   103 sec
Syracuse / 2 /    1.05500  /   111 sec
Zandvoort / 3 /    1.07220  / 96 sec
Spa 67  / 3 /    1.09300  /    227 sec
Bugatti / 4 /    1.01000  /   110 sec   
Silver /    4 /    1.00500  /   100 sec
Nurburgring / 5 /    1.01800  /    552 sec
Mosport /    5 /    1.04000  / 91 sec
MosportRain / 5 /    0.96000  / 97 sec   
Monza  / 6 /                          1.13000  / 99 sec
Oulton2- Gold Cup /   6 /    1.04200  /   101 sec   
WatkinsGlen  /    6 /    1.07500  /    74 sec   
Mexico  / 7 /    1.01000  /   123 sec   
Jarma  /   7 /    0.95500  / 97 sec
Kyalami 68  /   7 /    1.03200  / 91 sec
>
This should save you hours of testing.  It is possible that I may have a different version of one of these tracks from your version.  If you get unusual results during racing, consider retesting that track.  I have updated my tracks for AI and other  improvements when I find a relevant update ... this could possibly change one or more of the above listed values.  As an example ... Oulton has several different versions / improvements available.

The track67.ini values listed above are the key to my 60fps 1967 Historical set up; if you have any other questions let me know.

Best of luck and good racing,

Roger

Edited by Roger_F, Feb 10 2018 - 10:12 AM.