Information for Customizing AI drivers in AMS2

I would also like to know that. Although a potential inconsistency in lap times would also mean inconsistent races.

 

I put consistency very low - around 0.08, and happened that G.Berger with "McLaren" (classic formula gen3 - 1991), dropped often to some, 14th position... but S.Modena (skill 0.84, consistency 0.17) finished once 3rd, 4th. So lowering consistency in 0.10 - 0.35 range makes races more diverse, in the absence of mechanical AI failures (so Minardi can take 6th place).
In this case winners still be Mansell or Patrese, Or Senna, etc.... because not all leaders will have this skill drop in one race, so backmarkers should not win.

 

Just seen this video online. Seems well worth a watch.

 

The video made me realize how exciting this custom AI thing has been for those of us who love to over-analyze any set of data.

 

Yes, it seems essential for anyone who uses custom AI. Some good suggestions and requests, too.

 

I kept on tweaking the AI file for the Fiat Uno. Here is what I did:
I put everything to default I did not have information about, but tweaked Race skill and consistency more heavily. Qualifying skill is the same as race skill.
So the overall results at the end of the championship still largely determine the race skill. However, every time a driver placed 5 places or more better than in the end results, I increased his race skill a bit but he would get a low consistency score. For example, a driver would end in 8th place overall but but was in third place two times. He would get 0.03 to his race skill but a 0.2 consistency.
I don't quite know by how much racing skill is reduced, when the consistency value does its thing so that was a bit of guesswork. I also assumed that a consistency of 0.2 mean that 8 out of ten times, the skill would be reduced, but this is only guesswork on my part and could be completely wrong. I would also like to know what the default value for consistency is.
What do you think about the approach?

 

only bought ams2 in the recent sale and, as primarily a single player, the custom ai intigued me. i've started to play around a little and so far i've only adjusted skill and qualy values leaving the rest to whatever their default values are

setting up a gt3 race v my custom ai at the same ai strength the ai lap times dip by 3 seconds or so in qualy compared to the default ai times and i don't get why. i assigned random values to skill level between 0.2 and 1.0 and asssigned qualy skill between +/- 0.08 of the drivers skill level. what is it i'm not understanding?

 

After the experience of AMS1, where the open wheel drivers were less consistent than the tin tops, I chose the Ultima and the 2019 Stock Car, both single make classes, to better to observe differences in lap times. Both classes have quite tight fields in my own experience of racing them.

I gave one third of the field a consistency of 1.0 (and gave them first name of "Hi"), one third a consistency of 0.5 (first name "Mid") and the rest a consistency of 0.0 (first name "Lo").
All the other values I made 0.98 leaving just a tiny bit of room for...well, I'd already done a lot of tesing with values of 1.0.

At many circuits the field spread was only a few tenths (Imola, Hockenheim, Interlagos current and Historic, Suzuka). Some were a little wider (Nurburgring GT and Spa, around a half second), and Curvelo was the widest around eight tenths, the Ultima tended to be a narrower field than the Stock Cars but most of time difference in the Stock Cars were exceptionally quick first laps by one or two drivers, the rest of the field were very close.

In all the races there were as many Lo drivers at the top of the grid as Hi drivers at the bottom. The field spread was fairly consistent lap to lap, blockages included but not pit-stop laps. The whole field stayed within a couple of tenths, all the 'Last Lap' times vs the single fastest time.

If the consistency were to have a meaningful effect I suspect it would have been fairly obvious in lap time over a number of different tracks. I have found the same with all the other attributes I've tested so far and maintain that at present the race_skill and qualifying_skill are the only values to have any meaningful variation.

The current inability of the AI to overtake each other even as much as in a modern F1 field makes testing championships a bit pointless as well.

In the course of testing I found a couple of errors in my previous file that have been corrected. This one includes some extra columns that I make use of myself and offers a quick (for me) way of just updating a single class.

I've highlighted the duplicate Porsche Cup livery_name #911 in the 4.0 and 3.8 cars, and Marcel Gomes in the 2019 Stock Cars because he seems to be assigned to more than one livery. The only way I've found to prevent him entering twice (in a full grid) is to drive his car myself.

You don't have to use it but you might make use of the livery_name spellings.

 

thanks for that info. i assumed skill ratings were purely random between 0-100. now i'm getting to understand what's happening and why, i should be able to get thinks pretty quickly to my liking

 

warning- you will be replying to a clueless idiot

in the ai creation tool it says you can "re-roll attributes for all AI by going to "AI Proficiencies" in the menu above and click "Randomize!"

i can't for the life of me see any menu with "ai proficiences" in it. any directions would be appreciated

 

Hey Hal, have a look at the screenshot:

 

it's not there for me. nothing right of help.

 

When I go into the file, it takes about 4 seconds, before it appears. Do you view it in Google drive in your browser? You also want to make a copy of it.

 

First you have to make a copy of the original file and save it into your Google docs, so that you can edit it. Then you have to acknowledge and accept the Google warnings to enable the macro that does the randomizing. After that it works as @Blaine7 describes.

 

Also @Hal you may be using an older version of the tool...?

 

i downloaded it today from this forum. looks like i haven't done it right. i'll have another go later tonight or tomorrow. thanks for the help

 

ok got it now. i'll give it a whirl

ok got it now! thanks

 

just gave this a test run and it worked flawlessly. great job, love it. it saves so much time. thanks so much for sharing. it's really appreciated