Information for Customizing AI drivers in AMS2

Imo he didn't really give too much more info than what was said by @gian in this thread and did his testing about the same way as I did for GT1.
I would have loved to know other peoples results with consistency and stamina, as I couldn't feel any difference in my tests.
Also there is almost no variance (no matter the consistency) in qualifying. If a driver in the same car has a higher qualifying skill value, he will always qualify higher than the driver with the lower qualifying skill, except for traffic or slipstream. Even if it is 0.95 vs 0.945, he will always be that little bit faster, which makes it really boring.

 

Well, it was a useful summary.

 

AI are strictly behaving according to their skills. Can be boring, but I would rather see Senna in front in a historic F1 livery set than, let's say, Martini driving a Minardi. Still better than the random abilities the game offers without customized AI.
To have more unpredictable results, we would need to edit some kind of failure rates of the cars or the drivers.

 

I've done more testing on the Aggression and Defending attributes.

I set all drivers to a high value or a low one for the race_skill and qualifying_skill.
All the drivers with a high value (1.0) I set Aggression and Defending to 1.0.
All the drivers with a low value (0.8) I set Aggression and Defending to 0.0.

Nothing happened at Monza - none of that faster drivers caught the slower ones quickly enough. At Spa things changed. The faster drivers caught the slower ones in a lap or two but...

The top grid spots were occupied by an equal number of fast and slow drivers and some high value drivers were at the back of the grid. During qualifying low skill drivers with aggression and defending values of zero still managed to pass some of the drivers with high skill values.

During the race the lower skill drivers were passed and some passed back, frequently before Les Coombes or after Blanchimont. However, the lower skill drivers didn't catch up to the higher skill drivers at the front but fell back a little though they still squabbled with drivers of a higher skill behind them.

So far I have no indication of the Aggression or Defending values having any effect so it seems race_skill and qualifying_skill are still the only two having any noticeable effect.

The file I uploaded last night had errors. I updated the spreadsheet but forgot to update the zip file.
In this file all the livery_names have been checked in game.

For those using eSTee3's Template file I've listed the remaining differences in a tab delimited text file - I didn't know he'd updated his file until Harry H mentioned it and I can't find a way to reach via PM. One of the livery_names, Flavio Trindade, #34 in the Omega Stock cars is missing from that Template.

 

Is it possible to have custom AI with a dedicated server?

 

... link to your updated file, please?

 

No it's not

 

Harry H, I just re-uploaded it. Sorry about that.

 

OK I updated my copy of eStee's google sheet with your latest edits. He has a link to that and I expect he'll update his list shortly. Thanks!

EDIT: also added the missing Flavio Trindade, #34 in the Omega Stock cars

 

The only random value we have is 'consistency' as far as I know. Would be great to assign a value like 'engine reliability' to the drivers. Especially for the older cars this would level up the immersion quite a bit because back in the days half of the field had a DNF due to engine failures etc. Couple that with lots of smoke coming from the back of the cars and we have a 100% perfect experience.

 

That is 100% the opposite of my experience.

How long are the races you are running? I've been doing 30-45 minute races and am finding in many ways that the "secondary" values (aggression/defending/stamina/consistency) are mattering more than the skill & qualifying numbers.

The AI files I've been creating are kind of like your experiment--I've been using a narrower range for skill ( to create closer racing), but very wide ranges for the other values (just to see what happens).

I'm doing extensive GT4 & F-Classic-3 championships right now and my highest skill drivers are being dominated by people a little lower in skill but with other numbers that are higher.

 

The longest race I did was in the GT1 cars, documented a little while back, that was set for one hour but it sort of fell apart after 40 minutes when the AI started pitting for fuel and retiring shortly thereafter. I found no change after a few laps and didn't see the point of long races.

I've recently been using Stock Cars, Ultima, Ginetta G40s, probably the best three in the game for me right now, as well as F-V10s Gen 2, F-Reiza, F-Ultimate (without driving them myself).

I've avoided GT3s and GT4s as I figured they were more popular so others would be testing those and because of the AI overtaking issue I felt championships wouldn't be worth the bother plus the difference in chassis was a confounding factor.

When I assigned every driver the same race or qualifying skill I found there was still a good bit of random values being applied that lasted through qualifying and fairly short races

It's really good to see a different experience. I'll have another look with a wider range of cars.

 

Thanks Harry H.

I've been doing this stuff in Excel since messing with the hype and magic_grip in GPL's Driver.ini files. I can't use the google thing but had my spreadsheet set-up as soon as the custom AI was implemented.

 

Does anyone have AI files for Formula Trainer and Formula Trainer Advanced? I just can not get mine to work.

 

Do you mean driver names aren't working or attribute values or something else? I have all 67 files working for names and most for race or qualifying skills. I haven't tested karts for performance nor all of the Copas.

 

I used exactly the livery names as showing up in game menu, but none of my driver names is there when I start a race. So, if you have a working file for F-Trainer, could you please upload it here?

 

Does this work for you?

Are you using eSTee3's file or mine or are you hand crafting the XML perhaps?

 

Thx. Will give it a try when I am back at my PC. There must be some syntax error in my file.

In general, I just rename a working file of some class, then rename liveries and driver names (and later on the other values). Works perfectly, but here I must have messed something up.

EDIT: Works. Changed driver names, added country flags and all is fine. Must have deleted a " or > in my file coincidentally. Thank you!

 

But anyone tested if the situation change with 10 ai vs 25?

 

Hey all - lots of great stuff in this thread. I've been playing with 1991 F1 driver AI and it has sent me down a rabbit hole making spreadsheets (like some other brave/mad souls here).

It's not even directly applicable to the AMS2 AI parameters (yet at least) - but I am particularly interested in direct driver comparisons, i.e. ignoring the cars as much as possible. I've started comparing session lap times for team mates though the 1991 season and it is interesting, though I'm not convinced it is actually useful. My starting point was wondering if I if I built this up to a point where over multiple seasons you would have a lot of comparisons for different pairs of drivers (as they switch teams) to allow you to build a kind of relative ranking/metric of how quick they are, which you could use to inform setting AI drivers' skills.

Time-consuming madness, but I wonder does anyone know of anything like this already existing anywhere?