AI Brake Power vs AI Brake Grip

[RotaryRacer787b]

I've been doing some testing with the P&G mod and the two brake settings for AI to get them to brake more realistically. I have tried some of the settings via Von Dutch, and looked through the forum, but wonder if there is a better use case example for when to tweak each line.

For example, in a given situation when trying to improve AI braking into a corner, what line is better to tweak first? For me, I am seeing issues getting the AI to brake in slow corners where it looks as though they brake too early, then sort of pulse the brakes to get to the corner; like they slow down too much too early.

After fixing that, it seems I struggle to get them to brake in high speed areas early enough to keep them from running over me.

 

[jgf]

Those two settings in the plr file are global, they equally affect all AI in all cars in all mods; so obviously there is no one setting optimum for all (one reason many people have multiple installs of GTR2 so settings can be more appropriate for different type cars - GT, F1, prototypes, etc.).

Here is a little info from an old P&G thread which may help generically-
"If you patch your .PLR file pressing the PLAYER button in the PnG Launcher it makes these modifications to the AI values:

- AI Brake Power Usage="1.15000" // Fraction of theoretical brake power that AI attempt to use
- AI Brake Grip Usage="0.99000" // Fraction of theoretical brake grip that AI attempt to use
- AI Corner Grip Usage="0.92000" // Fraction of theoretical cornering grip that AI attempt to use

In my opinion, those values are excessive. With those values the AIs have more grip than your car so they can take the corners faster and most importantly they have more brake power than your car (and more grip when braking) so they start braking later than you (your car) can do, and they brake harder, so if you have an AI behind it will probably hit you. Also, it is almost impossible to outbrake and AI using a similar car than yours, which is annoying, because it feels like the AIs are cheating, well... they are actually cheating.

So after a lot of tests adjusting those values I finally found something that works better (for me, at least). These are the values that I personally use and I'm comfortable with:

- AI Brake Power Usage="0.90000"
- AI Brake Grip Usage="0.60000"
- AI Corner Grip Usage="0.85000"

Basically I tried to adjust the braking point of the AIs so it doesn't feel like they are cheating, and match it more or less with my braking point. I also reduced the grip of the AIs when taking corners and when accelerating so they are not excessively fast. For some reason this seems to work better during qualify. During races AIs are a bit slow at the beginning but they get faster after a few laps and make for a good challenge, I think.

\\\\\\\\\\\\\\\\\\\\\

Actually they don't have more cornering grip than you - 0.92 means 92% of theoretical maximum. This is actually slightly lower than the "as supplied" GTR2 figure of 0.93. The AI Brake grip useage is only 1% higher than the as supplied. The brake power usage is limited by the brake grip usage - you can have more brake power but if all it does is lock your wheels you won't brake any faster." (italics mine)

However it sounds like your problem involves editing all the talent files, in each is-
// Increase attempted low-speed cornering by adding a minimum onto calculated speed.
// Reduce attempted high-speed cornering by multiplying speed by a number less than 1.0.
// <adjusted speed> = CorneringAdd + (CorneringMult * <original speed>)
CorneringAdd = 3.454 // Higher number increases low speed cornering
CorneringMult = 0.887 // Higher number increases high speed cornering

Tom's Driver Shop and Tom's Talent Scout makes this easier. Note that the talent file (.rcd) name is irrelevant, the game reads the name on the first line of the file. The location of the talent files doesn't matter, the game will access them anywhere in the install (all the separate folders are just for us poor humans). And there should not be multiple files with the same driver name, so you must be creative with names if the same driver appears in different series - John Dough, J. Dough, J Dough, etc.

 

[IMC '67]

Here's my take on what the parameters actually do on track:

AI Brake Power Usage = How late the AI brakes.
AI Brake Grip Usage = How hard the AI brakes.

As for slow corners, the AI will always be a bit hesitant braking into a hairpin which is something that can't be totally fixed in my experience. It can be improved though by adjusting the CorneringAdd value in the AI talent .rcd files. (Affects attempted low-speed cornering by adding a minimum onto calculated speed). I use 1.2 for all the AI in my 1967 WSC game but I also use much higher AISens and AIGripMult .tyr file values than P&G. Another (more advanced) way to improve the AI slow corner behavour is to edit the .aiw waypoint parameters for a specific corner by optimizing the AI racing line as well as moving the collision and the cut track boundaries.

I wouldn't recommend using any advice from the Von Dutch thread as it's full of faulty crap. And there's only one real Von Dutch...

 

[jgf]

I wouldn't recommend using any advice from the Von Dutch thread as it's full of faulty crap

A scathing statement considering the extensive threads on NoGrip where dozens, perhaps hundreds, of people tried the settings, gave feedback, and helped von Dutch finalize all that work into a relatively simple tutorial.

There are no absolute best settings as the final result depends on the cars, tracks, and what someone expects of AI.

 

[IMC '67]

I haven't read the Von Dutch AI thread since it was an active thread on NoGrip but believe me, it just had too much advices based on faulty facts and/or inadequate testing, and I find it problematic that it is still refered to.

 

[RotaryRacer787b]

So after a lot of tests adjusting those values I finally found something that works better (for me, at least). These are the values that I personally use and I'm comfortable with:

- AI Brake Power Usage="0.90000"
- AI Brake Grip Usage="0.60000"
- AI Corner Grip Usage="0.85000"

Very helpful!! It seems as if my testing yielded very similar parameters for the AI grip and braking behavior as you, and my numbers are not far off.

However it sounds like your problem involves editing all the talent files, in each is-
// Increase attempted low-speed cornering by adding a minimum onto calculated speed.
// Reduce attempted high-speed cornering by multiplying speed by a number less than 1.0.
// <adjusted speed> = CorneringAdd + (CorneringMult * <original speed>)
CorneringAdd = 3.454 // Higher number increases low speed cornering
CorneringMult = 0.887 // Higher number increases high speed cornering

I think you are right. Looking at the talent files, some have values around 2.0 from the P&G cars, while others I am adding in are closer to 1.2 but have higher tire grip. This is why I would see some inconsistent values around changing just the brake power and brake grip (would fix some cars but not all).

AI Brake Power Usage = How late the AI brakes.
AI Brake Grip Usage = How hard the AI brakes.

As for slow corners, the AI will always be a bit hesitant braking into a hairpin which is something that can't be totally fixed in my experience. It can be improved though by adjusting the CorneringAdd value in the AI talent .rcd files. (Affects attempted low-speed cornering by adding a minimum onto calculated speed). I use 1.2 for all the AI in my 1967 WSC game but I also use much higher AISens and AIGripMult .tyr file values than P&G.

I have not seen this explanation before and it makes sense. Also, I think where I am having issues is trying to cobble together multiple cars from multiple sources to try and get a 1966 grid. This means some of the cars yield the tire settings you have, and a corresponding talent to match, while others have a lower tire grip, but higher talent cornering value to offset it.

 

[jgf]

full of faulty crap

faulty facts and/or inadequate testing

problematic that it is still refered to.

And what is your reference for this statement? "faulty" "faulty" is just opinion. You consider over two years of work by numerous beta testers to be inadequate?

It is still referred to because it works. Have you something better?

 

[IMC '67]

You consider over two years of work by numerous beta testers to be inadequate?

Yes, GTR2 has a fair amount of hardcoded randomness so if not putting in enough testing time will lead to that the testing becomes inadequate which will result in misleading feedback. A typical tester of a new setting would only do a couple of 10 lap races and then post his "report" in the thread. Also, tester volunteers are often not really qualified for any serious testing although there are of course exceptions.

It is still referred to because it works.

Several advices on important AI topics in the Von Dutch AI thread didn't work (i.e. the parameter is not used by GTR2) or was incorrect (e.g. the advice said to increase a value to achieve something when you in fact should decrease it).

Have you something better?

Yes, check out the AI related posts in my game thread. Here are some recent posts to get you started:

The International Manufacturers Championship '67

I'm pretty sure this is the actual spot in the game for the incident although the low wall here is only a sidewalk curb. Note also the identical diamond shaped pattern for the house window as well as the "NINO" writing on the wall.

www.racedepartment.com

The International Manufacturers Championship '67

I'm pretty sure this is the actual spot in the game for the incident although the low wall here is only a sidewalk curb. Note also the identical diamond shaped pattern for the house window as well as the "NINO" writing on the wall.

www.racedepartment.com

The International Manufacturers Championship '67

I'm pretty sure this is the actual spot in the game for the incident although the low wall here is only a sidewalk curb. Note also the identical diamond shaped pattern for the house window as well as the "NINO" writing on the wall.

www.racedepartment.com

The International Manufacturers Championship '67

I'm pretty sure this is the actual spot in the game for the incident although the low wall here is only a sidewalk curb. Note also the identical diamond shaped pattern for the house window as well as the "NINO" writing on the wall.

www.racedepartment.com

The International Manufacturers Championship '67

I'm pretty sure this is the actual spot in the game for the incident although the low wall here is only a sidewalk curb. Note also the identical diamond shaped pattern for the house window as well as the "NINO" writing on the wall.

www.racedepartment.com

 

[Speednut357]

The Ai braking in the Get Real Physics mod seems to me to be the most realistic. And looking closely at what he did in that mod it looks like changing the tire slip curve for each car was the main factor.
Not that any of us have the skills or time to do such a thing, but it is easy to pinch his tire files. Personally if I want a good race against Ai I use the porsche with that mod and do a Porsche cup style race.

 

[jgf]

A typical tester of a new setting would only do a couple of 10 lap races and then post his "report" in the thread.

And you know this how? True, some may have done this, but many others ran numerous races closely watching AI performance. Try not to generalize too much.

Also, tester volunteers are often not really qualified for any serious testing although there are of course exceptions.

The same people who populate forums with "expert" opinions on how accurately a cars physics are modeled when they have never even driven a high performance car, much less a race car. Again, do not generalize too much.

Several advices on important AI topics in the Von Dutch AI thread didn't work (i.e. the parameter is not used by GTR2) or was incorrect (e.g. the advice said to increase a value to achieve something when you in fact should decrease it).

And all of this was discussed, and corrected, in the threads referenced.

check out the AI related posts in my game thread

And by what parameters should your work be considered more accurate than that of von Dutch? Because you say so?

Over the past fifteen years many people have used von Dutch' work, either as-is or as a basis for their own continued tweaking. Now one person steps up and declares all this to be "faulty" and "crap", without any corroborating evidence. What arcane information have you discovered? What new data have you gleaned from the game? I'm not saying you are wrong (other than your assessment of "crap", which is not valid critique), only that you must convince us you are right ...and thus everyone else was wrong.

 

[IMC '67]

And you know this how? True, some may have done this, but many others ran numerous races closely watching AI performance. Try not to generalize too much.

Because I followed the thread closely and remember getting upset almost everytime someone posted feedback. A typical tester would say something like "I just did a couple of 10 lap races and found that ...".

The same people who populate forums with "expert" opinions on how accurately a cars physics are modeled when they have never even driven a high performance car, much less a race car. Again, do not generalize too much.

With "qualified" for testing I also mean being ready to spend the time needed along with not being retarded.

And all of this was discussed, and corrected, in the threads referenced.

You continue talking about multiple threads, I remember it as one sticky thread at the top. And I can't believe most of the errors were corrected.

And by what parameters should your work be considered more accurate than that of von Dutch? Because you say so?

Yes, because I have put in the time needed for testing/playing to verify my discoveries. You would be amazed to know of how many GTR2 file parameters that are not used by the game that people still are discussing as if they were used.

Now one person steps up and declares all this to be "faulty" and "crap"

I shouldn't have written my original post that way as not every advice in the thread was crap but the problem is if you can't know what's valid or not, things can get very frustrating for anyone who is serious about improving the AI in their game.

What arcane information have you discovered? What new data have you gleaned from the game? I'm not saying you are wrong (other than your assessment of "crap", which is not valid critique), only that you must convince us you are right ...and thus everyone else was wrong.

Again, check out the AI related posts in my thread and give my findings a try to see for yourself.

 

[RotaryRacer787b]

Based on not being able to have talent files with the same name, is there logic anywhere to when the game searches first? Using the talent tool, I noticed I do have some duplicate from the various mods and I’m wondering if this also caused part of the issue as both have different values.