How to adjust AI/Player car acceleration at the start of race?

Posted by Noog 
Did some testing a while ago concerning the very start of the race, ie the first few metres after the lights. I set all cars to equal power/grip values in the PF and went straight to race start.

It seemed to me that the lower the power setting (equal for all cars), the slower I was relative to the other cars at the race start, whereas with a higher power setting the faster I was relative to the CC cars. I think I identified the rough mid point too, ie where I was the same as the CC cars, but I forgot to write anything down.

The thing is, some mods seem to have dealt with this but others don't. So some mods are very realistic, but others let me charge from the back faster than Ayrton Senna on a rainy day at Donington!

What's the best/easiest way to deal with this issue as I'm pretty sure the PF isn't the place to do it and I don't think the answer is likely to lie in the MD settings either.

I did have a brief look at physics and torque settings but I was wondering if a modder can shed any light?

Edit: adjusted title to read acceleration rather than speed



Edited 1 time(s). Last edit at 10/02/2019 05:38PM by Noog.
No takers eh? ;-)

Oh well. Here's what I THINK I've learned over the last 48hrs. I tried to be as as concise as I could and I hope it's of use to someone.

In relation to the OP, the problem is as much to do with grip as it is power.

A player car with a power rating equal to all other cars but with low grip rating will accelerate more slowly off the line than the others.

That's logical, right?

OK, let's give all cars a power value of 750. To the player, the car will feel good on the track with a grip of 16000+, but with a grip of 13000 it will feel uncontrollable. At the start, with a grip value of 16000 or thereabouts, the player will accelerate from the line at about the same speed as the better CC cars.

(The actual figures I'm using here are very rough and off the top of my head, but broadly speaking they are indicative of what I'm getting at).

On the other hand, a car with a power value of 525 will feel just as good (or even better) on the track with a much lower grip value of say 14000+; but it too will become slippery if the grip is turned down to 11000.

So, imagine you had a mod where the power was quite low, e.g. 525, but the grip values were generally higher than necessary, say in the range 15000-16000. In this example the player car will struggle to keep up off the line.

Conversely, a mod with power set at 950 but with the same 15000-16000 grip values is going to make the player car artificially quick immediately after the lights go green.

It's actually quite simple isn't it?

Be aware that I haven't tested this properly yet so it might not be true! But based on my experience of having started lots of short races spread over all the pre-2001 mods I'm pretty sure that's what I've been seeing.

One of the implications of this is that any PF and therefore Performance Calculator, such as the one I've been working on for the last two years, has to build in some sort of inversely related power versus grip function, (ie raising the range of grip for all cars as the power increases and lowering the range of grip for all cars as the power is lowered), which needs to use the player car versus CC car 'speed off the line' phenomenon as its grounding reference - and this in turn will make all mods easily tuned to offer a consistent and more realistic experience at the start of a race.

Based on what I HAVE tested so far I'm confident that such a feature will completely eliminate the problem identified in the OP - and in essence the solution outlined here is (I believe) exactly what some of the modders have realized (and others have not).

That's not to criticize anyone of course; some of the mods are very old and I'm grateful for all of them, but if what I've said above makes any sense at all, it's something probably worth bearing in mind if, like me, you hope to bring a new mod to the community one day. Similarly, there are several mods which play MUCH better with a decent set of PFs.

Any thoughts?



Edited 1 time(s). Last edit at 10/04/2019 02:47PM by Noog.
Certainly seems like you're on to something. I guess the next step is to go to the extreme ends of the ranges and really confirm whether there is something in these values that causes a Player vs AI difference.

Ideally I'd want to engineer a situation where it's actually harder for the Player off the line than in the original game, because the AI can be over-cautious into T1 and there's always an opportunity for the Player to just gun it down the inside and make up loads of places. If we can give the AI an advantage off the line it could negate that and make for a much better Lap 1 experience.

=====================================================


Intel NUC 8i3, 8GB RAM, MS Sidewinder Wheel
TomMK Wrote:
-------------------------------------------------------
> Certainly seems like you're on to something.

Cheers Tom.

> guess the next step is to go to the extreme ends
> of the ranges and really confirm whether there is
> something in these values that causes a Player vs
> AI difference.

Have been testing a low power mod (520) based on my home-brew 1978 mod since I wrote the last entry. (It's really a mix of 75 cars with their air-boxes removed and reworked front ends, along with some 79 and a couple of 77 cars, but as you know, that doesn't really matter in this instance). Anyway, I originally had the grip values ranging from about 16000 to 14000 (based on habit more than anything) and I reduced the middle of the car sets grip range until the start felt right, which turned out to be about 14000. Now, the start is now pretty perfect and the racing is as good as ever. However, as you said:

> Ideally I'd want to engineer a situation where
> it's actually harder for the Player off the line
> than in the original game, because the AI can be
> over-cautious into T1 and there's always an
> opportunity for the Player to just gun it down the
> inside and make up loads of places. If we can give
> the AI an advantage off the line it could negate
> that and make for a much better Lap 1 experience.

And yes, I agree completely and will definitely be trying to deal with that once I crack the main nut. So there's that...

However (again).

Using the same 520 mod, I've kept the grip value for CCs at the same 14000 Q&R, again with no variability or failures - and tomorrow I'll be testing four cars rated like so:

Andretti: 13750
Peterson: 13725
Laffite: 13700
Jabouille: 13675

Why such tiny gaps?

Well, I started off testing with:

Andretti 16000
Peterson 15000
Laffite 13000
Jabouille 12000

but Andretti and Peterson were way too quick (Andretti took pole with 1.4s to spare and Peterson took 2nd with 1.6s between him and 3rd placed Regazzoni's CC car) while Jabouille in particular was disastrously out of touch (8s off the pace), so each test I've reduced the gaps and reduced all the player cars relative to the CC cars).

Problem is, by the time I got to my last test I had brought it down to what I thought would be negligible differences (still with all CC cars set to 14000:

Andretti 13960
Peterson 13860
Laffite 13760
Jabouille 13660

Notice that all player cars are set lower than the CC cars in this test, but I'm STILL qualifying the top three in the top three, which I can remedy using the MD file later so that's OK - but what's really weird, is that Jabouille only managed 16th!

I wasn't expecting that - and it wasn't that I drove much differently - but I could definitely feel that his car was just a tad more slippery than the others. I guess it's one of GP4s hidden variables coming into play.

However (one more time), in the actual race my four drivers scored the four fastest laps, albeit not by much (0.1s, 0.1, 0.2 & 0.3), so I'm going to repeat that particular test tomorrow to see what happens. I must admit I'm a bit surprised at what I'm finding so far. The grip setting seems to be FAR more influential even in terms of minor changes than I had been assuming.

All that said though, it DOES seem to be an effective way to sort out those first few metres of the race - which ultimately is what I was aiming for - but as always, GP4 seems to find a way to create six more problems for every one resolved. I had a girlfriend like that once.
A quick update:

Here's my latest test.

All CC cars, 14000, 520, no failure, no variability.
All player cars 520, no failure, no variability.

Player car grip:
Andrett: 13750
Peterson: 13725
Laffite: 13700
Jabouille: 13675

Judging by yesterday's results I expected Andretti to be first, or near to it, with Peterson 7th or thereabouts, Laffite 14th and Jabouille last(ish).

In qualifying it was close. Andretti was 1st (just) but Laffite was 9th, Peterson 11th and Jabouille 16th. OK, I'll call it driver error with Peterson probably. But what about the fastest race laps:

Nope. An unexpected shambles.

1 Peterson set the fastest lap
2 Laffite 0.18s slower than Peterson
3 Andretti 0.26s slower
4 Jabouille 0.36s slower
5 Regazzoni (CC) 0.49s slower

How can this be? A different set of mathematical rules for qualifying and race? Oh happy days. The problem got twice as big. What's worse, it wasn't ME who set those times - it was the AI in between my player turns that set them in my so-called de-tuned cars. All my times were about 0.6s slower (ie pretty much midfield).

It's a tough nut with many layers....

I think I'll fire up my '98 mod and see if I can find the midpoint in that one, i.e. the 14000 as it seems to be in my '78 mod. Maybe that'll help illuminate this thing a little.
What length races are you doing with these tests?

=====================================================


Intel NUC 8i3, 8GB RAM, MS Sidewinder Wheel
(Before I start, apologies for long post, but I think you might find it interesting).

Generally I only race 17% races Tom. I know that skews things a bit, but I have to sleep etc and I'm hoping that when (if) I can pin the mathematics down it's easy to extend the principles out to 100% races.

I did make some progress yesterday. On the basis of the 520/140000 midpoint giving a nice balanced start line dash in my '78 mod I put that info back into my spreadsheet and created a new set of PFs based on the real world F1 results but anchored around that 520/14000 midpoint which informs what I call the four 'range anchors' in my spreadsheet, ie max power, min power, max grip, min grip. Then I ran a race.

Several incidents occurred in the second half of the race (because by now I had re-injected failures and variabilities back in the PFs - actually I forgot to take them out), but until the midpoint things ran exactly according plan. Indeed for more than one lap my four cars were in exactly the positions that I had predicted they would be and they looked set to stay there. Andretti's car felt nice but not ridiculously solid to drive and Jabouille felt OK, ie it was noticeably harder to drive than Andretti without being impossible. Overall, so far, I'm quite pleased.

TBH I'm a bit tired now and there's a large backlog of non-GP4 things I should be doing before tomorrow but what I'm planning to do next is to load up the '98 mod and repeat what I did above until I find the appropriate grip midpoint for the start in that mod. It should be easier this time. Then, once that's done, it'll be a simple task to compare the '78 & '98 outcomes and give my spreadsheet a set of rules to automatically assign the appropriate grip midpoint to all the mods in between.

Remember that I am working on the basis of gradually increasing the power by a factor of somewhere around 1.5% each year to give a linear progression, with a similar in concept but different multiplier being applied to the grip (which currently has a progression of about 0.7%/yr). Maybe one day I'll be able to build in quirks like rules changes etc to account for the years when the progression wasn't so linear, but certainly for now I just want each mod to drive and feel like it's part of a coherent set of mods that offer a strong sense of progression through the years - and most of all I want to resolve the 'too fast at the start line' flaw that affects so many mods.

The figures above and ever changing, but in 78 (so far), Mr Average will be getting default power of 520 and a grip of 14000 or thereabouts (it actually ends up being a bit less in the final PF), whereas in the 98 mod I'm expecting Mr Average to end up with something like, say, 700/16300, which would put 2004 (the newest year I've ever looked at) at about 770/17000. I pretty much have all this done at this point and it definitely generates some very refined results, but obviously there's a lot more work to really pin it down and the spreadsheet remains (and probably always will remain), a horribly complicated beast that only I could understand. Seriously! Even if I leave it alone for a couple of days I have to spend an hour or more re-learning how it works, as values are influenced by one set of modifiers before being changed by others etc etc. It's like a bowl of spaghetti and probably massively over-engineered TBH, but it's become an addiction (and I've always been a spreadsheet fetishist in any case!).

There is one big problem which I have uncovered however and I'm wondering if you can shed some light. In each race I did yesterday when the AI took control of one of my (effectively de-tuned) cars (520/137xx), it drove it quicker than me AND quicker than it was able to perform in the 520/14000 CC default ones. I just don't understand that, do you? It's not a showstopper but it's certainly an issue I hadn't really noticed before and a big mystery to me at this point - and currently I have no idea how to resolve it - although there will certainly be a way to compensate or hide it in my spreadsheet in time. Ideally though, I'd like to know how that particular behaviour is governed. Why does the AI do better when it's standing in for me than it does in it's own cars? It's like I'm missing an unknown variable somewhere. I'll continue to work around it for now but it would be nice to root it out. I'm thinking it might be something in the MD file but I'm trying not to bring them into my focus at this point.

I'm not sure if I've mentioned this on the board (because it's a bit controversial) but my current thinking is that PFs and MDs should be considered to have two distinct and NON-overlapping functions:

1. PFs govern the performance of cars and drivers relative to each other. They have nothing to do with re-creating real world lap times. They are simply there to create a range of power/grip values to regulate the spread of laptimes through the field. (And regulate failures and other stuff, but that's not relevant here). The main point here is that RANGE that they create.

2. MDs are entirely circuit oriented and have nothing to do with driver/car/team performance (relative to other driver/car/teams). The MDs are there to fit the RANGES created by the PF into the track behaviour as defined by the MD (ie using the track grip and grip factor values, predominantly) , ie this is how and where actual real-world laptimes should be simulated.

Anything else and we're stuck with the start line acceleration issue. That's my thinking.

Now, as I see it Tom, you have a really good feel for MDs and I know you must have laid out a ton of effort creating the ones you have. However, I'm hoping that if I can make PF creation a more or less standardized procedure it might then be possible to simplify the creation of MDs too. Eventually, as a result of what I'm trying to do, we might end up with a working model whereby when you create an MD for one particular track year, you will be able to generate modified MDs for all other track years automatically and simultaneously. I'm not sure how yet, but that's the idea. You'll still need to make adjustments for year specific anomalies, such as rain chance or the track being resurfaced in such and such a year etc, but the core factors of track grip, CC grip and default gearing could be informed by the power/grip ratio to be found in the PF - then voila, the start line acceleration issue can be put to bed and we can all get back to some great racing!

That's ultimately where I'd like us to arrive. It may well be (in fact it probably is to some extent) an unrealistic goal, but for me personally, I do get tired of editing and re-editing MDs (especially the CC grip factor values) over and over as I refine my performance calculator and change the PF values.

Just for context, let me remind you that currently my PF calculator automatically generates a PF based on real world historical F1 results for each race* (along with an incidental generic year average PF like a lot of the older mods have). Once I enter the historical results set for a given season it takes about ten minutes to arrive at a full set of PFs, which is the time takes to copy and paste the PFs into the 17 or however many PF files. Unlike most calculators it takes into account Q times and FL times but ALSO position achieved, as well as failures and variability. There are other things too (the secret sauce), but that's the core of it. When I look at my database of results (which increases rapidly owing to my tendency to do short races) it's very accurate. (It easily expands accurately to 100% races by the way). It's not that accurate PFs haven't been made before, there are plenty out there, (especially after '94), but it's the idea that we can maintain a consistency of feel, (esp. at the start line), between each mod year which I think might be unique.

Maybe at some point in the future we might work together on this Tom? How about we identify a couple of older mods around the same period with defaults we don't care for - and create something to really bring them forward as a 'matched pair', you might say. It's not like we'd need to bother with permissions as it would just be a few PFs and MDs.

Maybe most won't care or even notice the difference, I don't know. But I'm sure there are plenty of others out there who would appreciate it.

Finally, I guess I should add, what I'm doing is in no way intended to overrule or over-ride anyone else's efforts in this arena and I'm not trying to cast myself as some kind of expert either. I'm not. I'm just an obsessive tinkerer doing what I can to improve the GP4 experience for myself, primarily - and ultimately if anything good can be passed on into the community it's just a bonus and freely given.

* To be precise, the real world results are modified slightly, in that extremely bad Q and FL results are modified fit within a range I call "max Q and Max FL". This is because if you let a 5 min FL time through when everyone else managed to get round in 1 min it completely ruins the whole PF range spread. In other words if the real world FL was 1:02.00s and the 21st FL was 1:03.00s, the driver who recorded a FL of 5 minutes (because he, say, got a puncture on the second lap and retired shortly after) is modified to read 1:04.00s. All rough figures obviously, but indicative. Apart from eliminating anomalies like that, it's all based on historical real world results. Make sense?
I wouldn't be surprised if there's some weird legacy behaviour in the "multi-driver" mode you're using that isn't patched by GPxPatch - it's not a commonly used mode I think. Maybe something Rene can take a look at if he sees this. But I would conduct your experiments using single player mode with GPxPatch set to the "new behaviour", it's a "known good" testing environment.

So for me the most interesting point in all this is the start-line varience between Player vs AI which - if we can figure out how it works - could help the lap 1 experience nicely. I'll probably run a few tests on this myself shortly.

Regarding Magic Data in general, you probably already know I am always keen to nerd-out on the topic! I already have something similar to the system you mention where I create a baseline MD file (I use 2001 season because, well, obviously) and then can create a season-tailored versions (e.g. 2018) based on some year-specific modifications, e.g. Tyre Wear in the Pirelli-era years, or Fuel Consumption in the Turbo-era years, etc. It's far from perfect, but I think they're good enough to include as part of my GP4 Central mods. I don't know how widely they're used!

Would definitely be happy to collaborate further - check your PMs...

=====================================================


Intel NUC 8i3, 8GB RAM, MS Sidewinder Wheel
Sorry, only registered users may post in this forum.

Click here to login

Maintainer: mortal, stephan | Design: stephan, Lo2k | Moderatoren: mortal, Vader, stephan | Downloads: Lo2k | Supported by: Atlassian Experts Berlin | Forum Rules | Policy