AI Difficulty Calculator

[NinjaElmo 21]

@Gonira Interesting stuff, it looks like we had similar methods. My calculator is pretty much doing a similar thing, just mathematically instead. How did you get the top 10% times? It looks like it will take a LONG time to scroll down that far in the time trial leaderboards...I am interested in trying the average times for my calculator and seeing what results it spits out. If you've got the times on a sheet, can you send them my way?

Off the bat it looks like your Austria suggestion could cause issues, the AI seems particularly slow around here, even with strict corner cutting. I agree with you on Canada being an issue for that though. I honestly think the reason Monacco, Hungary, and Singapore have such extreme recommendations in my calculator is because they are essentially the 'hardest' tracks, so CM has slowed the AI down here to try to give people a better race. The base difficulty I have been using in my team is 85, for Singapore I put it on 95, and genuinely had the most evenly matched race yet, where I had to race right near my limit, and could only make an overtake when the AI made a mistake.

So far in my my team save I've been going roughly off the suggestions from my calculator, but unfortunately things seem a bit too easy. I'm placing 7th most of the time in my first season (ie. first behind the Mercs, Redbulls, and Ferraris). 

Edited September 16, 2020 by NinjaElmo

[NinjaElmo 21]

Also Canada in your results could possibly cause issues, but it could be down to that last chicane. In my calculator it's a +0 track, but all the feedback I have recieved on my calculator has been that that's too slow (2x much too slow, 2x too slow, 1x accurate). (As a side note: Austria on my calculator is a +4, but the feedback I have recieved is that it's still too slow (4x much too slow, 1x too slow). Let me know how Canada goes on strict settings with your recommendations though, people might just be flying over that last chicane and then complaining that the AI is too slow.

For comparison, here are our results side by side. I've bolded and underlined the tracks with biggest differences. I'd be interested on anyone's subjective thoughts on these, specifically Australia, Vietnam, and Abu Dhabi.

Track	Elmo	Gonira
Australia	+4	-6
Bahrain	-7	-4
Vietnam	+1	-7
China	+2	+2
Netherlands	+2	0
Spain	0	-5
Monaco	+17	-6
Azerbaijan	+1	-2
Canada	0	-7
France	+1	-1
Austria	+4	+1
Britain	-7	-7
Hungary	+11	0
Belgium	-5	-5
Italy	-1	-4
Singapore	+17	+4
Russia	-3	-3
Japan	-2	-5
USA	-2	+2
Mexico	+5	0
Brazil	0	0
Abu Dhabi	+4	-8

So there are some tracks that come out the same, some that have big differences. I think Monacco, Hungary, and Singapore are genuinely a pain to recommend a difficulty on, because I have seen so much conflict between people saying the tracks are too easy or too hard. As in plenty of 'Singapore sux the AI there are so slow!!!' and plenty of 'Singapore is STUPID the AI there are WAY too fast!!1!'. But I think the extreme recommendations from my calculator are a result of CM intentionally slowing the AI down on these much more difficult tracks.

Anyway, I'm open to critisicm and changes to my calculator, it seems like lots of people are using it off Steam so I want it to be accurate. I should be able to make some changes to it in a few weeks to make it more accurate.

 

[MilligramSmile 19]

@NinjaElmo 

do you suggest changing the difficulty before each race so that the forces are equal ?
And explain, what is the connection between complexity and strict cutting of corners ?

I used 100% base difficulty for the first 13 races. Right now I set 102% using strict corner cuts.

Edited September 17, 2020 by MilligramSmile

[NinjaElmo 21]

@MilligramSmile For now, I suggest only changing the difficulty of the following tracks: Bahrain -5, China +2, Austria +3, Britain -7, Belgium -4, Russia -3, Singapore +5 Japan -3. That's based on areas that my and Gonira's findings agree. Other tracks leave at your base difficulty. 

For the strict corner cutting conversation - basically some tracks can be much easier if the player cuts corners, because the AI never cuts corners. The last chicane on Canada is a great example, where a player can blast through on lenient corner cutting settings and reduce their lap times by 0.5s, whereas someone on strict corner cutting can't do that. But if you're using strict corner cutting, you don't need to worry about it, because you are playing by the same 'rules' as the AI.

As a side note, from my experience, some tracks have more difficult practice programs, particularly the qualifying program. I plan to work out definitively which tracks this happens on in the future, but if you are finding the qualifying program crazy hard to beat, it won't necessarily mean you will do poorly in the actual qualifying session.

[MilligramSmile 19]

the complexity of the tuning is so crooked that the hair stands on end ... I want to spit.

Watching videos on YouTube "my career" mode. The man has pumped the whole team to the maximum, he sets the difficulty to 110% and rides barely getting into the top 10, but his partner brings the Mercedes for more than 0.5 seconds...
As it is, the team is half a second faster than a Mercedes, but I fight at the level of McLaren, Renault, racing point ... although I have pumped out to the maximum of the chassis and aerodynamics... and it's hard for me to overtake them...

The most curve is the complexity and the closer to 110% - the more curve.

upd. also a partner who brings a Mercedes for 1 second in a race is the cheapest pilot on the market ... as well as the cheapest and not a wholesale pilot makes such pilots as Bottas and Hemilton to smithereens ... hahaha

Edited September 17, 2020 by MilligramSmile

[NinjaElmo 21]

From what I could tell, the difficulty jump starts after 105.
95 - 105 difficulty was around -1 second per lap, but 105 - 110 was also -1 second per lap.

My Team mode definitely has a lot of problems for CM to fix before players can get a consistent and challenging season without making changes and self-imposing rules.

[MilligramSmile 19]

we go further, start the 15th stage Italy, My team.
Loading C1 (fast circle task):
100% - 1:21.743
101% - 1:21.636
102% - 1:21.515
103% - 1:21.386
104% - 1:21.221

The total difference between 100% and 104% is 0.522 seconds
In Belgium C1 the difference between 100% and 104% was 0.968 seconds.
What the heck...

Edited September 17, 2020 by MilligramSmile

[MilligramSmile 19]

Previously, I could not get into the top 10 on software, but now I am the top 1 based on the results of C1.

Previously, I could not complete all tasks perfectly, now I am doing everything...

And most importantly, look at what I'm doing - the engine and stuff are all worn out...

My team is ranked 10th out of 11 in terms of power...

There is no development at all for my team...

I've already raised the difficulty from 100% to 102%...
What would be the feeling that my team is a beginner in F1 and take the last positions as it should be for beginners, I need to bet 110% ???

And yes, for 15 races I have not yet received any fines for changing the engine and elements... and AI already receive fines from the middle of the season, but they get few...

The first 11 races were interesting, I was able to take points only in 4 races, but then it’s boring... on a weak car I drive well or AIs drive badly... interest disappears...

Edited September 17, 2020 by MilligramSmile

[Gonira 21]

15 hours ago, NinjaElmo said:

@Gonira Interesting stuff, it looks like we had similar methods. My calculator is pretty much doing a similar thing, just mathematically instead. How did you get the top 10% times? It looks like it will take a LONG time to scroll down that far in the time trial leaderboards...I am interested in trying the average times for my calculator and seeing what results it spits out. If you've got the times on a sheet, can you send them my way?

To get the top 10% times, I scrolled. There's no other way. 3 to 5 minutes holding the down button for each track. A pain.

I Ideally wanted to go with a leaderboard time that would match my AI, to avoid these +2.whatever% multiplications, which can cause distortions, but no way I'd be able to scroll that far down.

These are them. Top 10% leaderboard cutoff times (dry weather), as in September 15:

Australia: 1:21.240
Bahrain: 1:26.395
Vietnam: 1:35.349
China: 1:30.391
Netherlands: 1:09.886
Spain: 1:15.857
Monaco: 1:10.924
Azerbaijan: 1:39.595
Canada: 1:09.487
France: 1:29.194
Austria: 1:03.986
Britain: 1:25.548
Hungary: 1:14.674
Belgium: 1:43.056
Italy: 1:19.856
Singapore: 1:35.645
Russia: 1:31.711
Japan: 1:27.469
USA: 1:31.466
Mexico: 1:15.072
Brazil: 1:07.782
Abu Dhabi: 1:35.319

[PTC 5]

 

 

 

Each driver with his difficulties on certain tracks. I am new to the game, my basic difficulty is 60. 
I play in My Team Mode and on some tracks I can be constant most of the time, but in others a total failure.
 Despite fulfilling all training programs very well (I get to be among the TOP 10), 
at the time of qualifying and race I am disputing the last positions with Willians. 
I really appreciate the work of you looking for the ideal AI, but for me, for now, I am not able to leave much of this base 60/65

Edited September 17, 2020 by PTC

[Gonira 21]

15 hours ago, NinjaElmo said:

Let me know how Canada goes on strict settings with your recommendations though, people might just be flying over that last chicane and then complaining that the AI is too slow.

Are you mad? Strict is unplayable haha, I'd be getting a time penalty every lap. I play on regular.

BTW, you asked about Australia. I tried it and to me AI pace seems right, if not a few tenths slower than it should.

I did some sessions just for difficulty checking. 20 or so time trial laps to relearn the track, then jumped into short quali + short race to get a feel of the AI.

Brazil, Australia and China felt about right to me at around 85. I'd say probably australia at 79 would be too easy? But that's when corner cutting comes. I think there's quite some time to be gained there at sector one if cutting rules aren't strict, and mine were not.

I did Spain also, and it was a pain, I was one second off the pace, but this is probably due to myself, I hate that track.

Edited September 17, 2020 by Gonira

[Gonira 21]

3 hours ago, MilligramSmile said:

I've already raised the difficulty from 100% to 102%...
What would be the feeling that my team is a beginner in F1 and take the last positions as it should be for beginners, I need to bet 110% ???

And yes, for 15 races I have not yet received any fines for changing the engine and elements... and AI already receive fines from the middle of the season, but they get few...

The first 11 races were interesting, I was able to take points only in 4 races, but then it’s boring... on a weak car I drive well or AIs drive badly... interest disappears...

You're good.

I'd probably go all the way to 110 if I was you. At least 105.

And I'd start racing on leagues. Be a legend.

[pacers101 19]

4 hours ago, Gonira said:

Are you mad? Strict is unplayable haha, I'd be getting a time penalty every lap. I play on regular.

BTW, you asked about Australia. I tried it and to me AI pace seems right, if not a few tenths slower than it should.

I did some sessions just for difficulty checking. 20 or so time trial laps to relearn the track, then jumped into short quali + short race to get a feel of the AI.

Brazil, Australia and China felt about right to me at around 85. I'd say probably australia at 79 would be too easy? But that's when corner cutting comes. I think there's quite some time to be gained there at sector one if cutting rules aren't strict, and mine were not.

I did Spain also, and it was a pain, I was one second off the pace, but this is probably due to myself, I hate that track.

I think a great point is being raised here which is that there is far too much difference between the corner cutting rules. Strict is virtually unplayable for the average player (I play AI level of 95-100 and there's no way I'm completing 25% race without multiple penalties... I've even seen guys like TRL Limitless struggle with strict rules). But then normal is just way too lax. On some tracks (Canada probably being a prime example) you can cut multiple corners with 4 wheels off the track, and not even get a warning.

And as you rightly say, this massively distorts the AI difficulty in some tracks.

[Gonira 21]

Yes. The swings caused by regular corner cutting rules would be possibly bigger than those caused by AI pace itself, so our model alone probably won't make much sense for most.

The game really needs an intermediate stringency level. Both the current options are too extreme in opposite directions, both are terrible.

[NinjaElmo 21]

@Gonira

7 hours ago, Gonira said:

To get the top 10% times, I scrolled. There's no other way. 3 to 5 minutes holding the down button for each track. A pain.

Oh boy that would have taken a long time, I salute the boring and hard work.

I plugged those times into my calculator to see how they affected the numbers. Dealing with the numbers is a mess, so you'll have to bear with my observations on the results so far:

- Using averages instead of world records made some of the recommendations closer to yours (n=8), some further away from yours (n=7), and some had negligible change (n=5). Austria came equal to yours, but from first hand experience, my calculator feedback, and general complaints I've seen on these forums, a +1 for Austria just isn't enough to speed the AI up to a competitive level.

- The extreme recommendations for my Singapore and Hungary suggestions were 'fixed', with Hungary now a +5 and Singapore now a +8. These still don't match your results exactly, but the difference is less extreme then they were.

- My Australia, Vietnam, Monacco, Canada, and Abu Dhabi recommendations all still very different to your suggestions. 

Here's some of my thoughts on why these tracks particularly might be causing issues with algorithmically calculating results.

Australia is usually the first track people time trial on, and has (or at least it did when I last checked) many many more times than on other tracks. This could throw off the 10% method, but my WR method suggests a +4, and boy do I hate Australia and think that it would be a track where you need to lower difficulty, not raise it. 

Vietnam, as a new track, lots of people might be wanting to try it out, increasing the number of time trial players and therefore potentially throwing off the 10% method.

Monaco, lots of people want to set a great time and show everyone how awesome they are. 

Abu Dhabi possibly has fewer time trialers because it's last on the list (? just a hunch, I haven't checked this). 

 

Anyway that's ALL speculation, just trying to find a story to the results so to speak. I think using a combination of our methods may give the best result, using WRs for some tracks, top 10% for other tracks, and just brute forcing it/finding a different method for other tracks that are causing issues. Where the nature of the baseline we are trying to use is changing (Australia, Monaco, and Abu Dhabi Time Trial leaderbaords, all for their different reasons), it might be better to just completely change our method and manually run through some different scenarios or something. Monaco particularly, because people seem to either love it (and complain it's far too easy) or hate it (and complain it's downright impossible). 

Anyway, that's my ramble so far. I'll look to update the calculator and defintiely use averages for Hungary and Singapore, since those seem most obviously improved by top 10% times.

[Gonira 21]

For my list, the fact there's more people or less people time trialing is not relevant, because my cut is a percentage.

A track that attracts a lot of newbies like Australia (since it's the first one) might shift the averages to the slower side though, but tbh I'm not sure.

I think the only failproof method would be a very consistent dude setting personal bests on each track and allowing himself the same level of cutting we'd allow (so it would need to be outside of time trials). Almost everyone is not that consistent though, it would have to be a pro. Some people post lists of difficulty per track based on their own performance. It could be a start if the guy is good, but inevitably there will be tracks where he is faster or slower.

And we doing it ourselves for our personal use is not something I'm in love with as well. If I'm underperforming in a track for my own fault, and not the AI's, I wouldn't want to solve it by pulling the difficulty back, but by pushing myself to improve.

This ain't easy at all.

[Gonira 21]

About Austria in my list, I wouldn't see it as a +1. Maybe a +4 or +5, if you consider the average (not saying it's enough though, this I don't know). My baseline of 85 was taken in Brazil, but in theory Brazil has one of the slowest AI's, so it isn't really a baseline.

[NinjaElmo 21]

Yep, were definitely on the same page about why we can't trust our own track performance/opinions. Unfortunately a perfectly consistent across all tracks person doesn't exist 😞 I was hoping with enough feedback on my survey I could essentially average things out, so essentially average out everyone's track inconsistencies, but I haven't got enough responses to get a clear picture. Seems like people are just using the calculator straight up and taking it as gospel that it will make every single track a competitive race for them.

I took down the youtuber Tom97's hotlap times in the hope he would give a consistent baseline, but his performance (even as someone who does a 'hotlap' series) varies track by track too. So even the pros have too much variance to provide a good baseline.

Then there's also that using different assist settings, damage settings, and rules, all makes certain tracks harder/easier. I use no assists, but people using full ABS and traction assist might find certain tracks much harder/easier than I do, which makes giving a consistent +/- by track recommendation difficult. 

[Gonira 21]

I did the same in 2018, checking a hotlap series of some youtuber - TRL Limitless, I think (we seem to have the same ideas, heh). Yeah, definitely there's variance. We could have a clearer picture though if we get this information from many of them and average the times. Actually, it would be very indicative of real life performance, but unfortunately, the number of guys who post an entire series of personal bests in every track is not that big. And then again, the problem of all these laps being under strict cutting rules. Getting numbers like these for regular cutting doesn't look like it's possible. There's the problem of them always using hotlap custom setups too, I'm not sure of how this would reflect performance against full race setups, or worse yet, default setups.

Edited September 18, 2020 by Gonira

[Gonira 21]

3 hours ago, Gonira said:

We could have a clearer picture though if we get this information from many of them and average the times.

So I just did that, but I only found one other youtuber apart from Tom97 who did this in all the tracks: Jaames. He is a bit faster than Tom in every track. Two guys = less variance at least. These are the average times between the two, fwiw.

Australia 1:19.021
Bahrain 1:24.955
Vietnam 1:32.917
China 1:29.055
Netherlands: 1:08.330
Spain 1:13.987
Monaco 1:07.918
Azerbaijan: 1:37.342
Canada 1:07.649
France 1:27.043
Austria 1:02.602
Britain 1:24.003
Hungary 1:12.696
Belgium 1:40.978
Italy 1:17.812
Singapore 1:32.738
Russia 1:29.751
Japan 1:25.487
USA 1:29.254
Mexico 1:12.927
Brazil 1:06.455
Abu Dhabi 1:32.194

Using this as reference instead of my previous choice of top 10%, the differences in recommended AI would be huge. China, Britain and Bahrain would be considerably slower. Singapore and Abu Dhabi would be much faster, and Monaco would be more than 1 second faster. All much closer to your chart, it seems.

No considerable change in Austria, though.

Edited September 18, 2020 by Gonira

[Gonira 21]

Ok, so I went for a different approach just out of curiosity, trying to eliminate as many variables as I could.

Again, using the leaderboards, but totally disregarding the top 100 laps, to get rid of cheaters.

From there, I searched the 3rd best time that would totally match my assists and setup situation (which is: TC on, ABS on, manual gears, no custom setup). 3rd best instead of 1st best just as a second layer of protection against cheaters.

At Interlagos, this gave me a 1:07.524. My 85 AI baseline lap is 1:09.607, which is 3.0854% slower.

I repeated this search in some other tracks and the results were surprising.

 

Track / 85% AI pole position grand prix lap / 3rd best outside of top 100 TT lap with assists same as mine and no custom setup * 1.030854

Australia / 1:22.557 / 1:22.691
China / 1:33.090 / 1:33.421
Spain / 1:17.478 / 1:17.586
Monaco / 1:12.227 / 1:12.410
Canada / 1:10.746 / 1:10.688
Austria / 1:05.789 / 1:05.534
Britain / 1:27.234 / 1:27.412
Singapore / 1:38.725 / 1:38.922
Brazil (baseline) / 1:09.607 / 1:09.607

 

This is WAY TOO CLOSE and really makes me feel like it's pointless to deviate from a fixed difficulty level. Fells like the difficulty levels were well tuned, but tuned for those using default setups and TC, perhaps?

Edited September 18, 2020 by Gonira

[stephen31416 23]

I really appreciate the effort put into this in the OP and elsewhere.  This was certainly a subject of interest to me in the 2019 game where I thought the AI was quite slow on some circuits compared to others.  That said....after making it half way through my first season I'm leaning to the side of 'it's not the AI that's inconsistent from track to track, it's the human' for the 2020 game.  Even Austria.....  

I found my 'base' level by spending plenty of time practicing at Spain and also tweaking setups so that I'm dialed in, and then setting up a grand prix weekend and finding the AI level that had me running at a very similar race pace (same compounds, very similar lap times in clean air) to my AI teammate--96 base level.  My very qualitative thoughts on AI pace after completing half a season (100% length, no assists, Renault w/ Ricciardo as teammate, all races at base level of 96 for AI) --

• I'm a bit faster than AI at China, Netherlands and Austria;
• My race pace is very similar to AI for Bahrain and Spain;
• I'm a bit slower at Australia, Azerbaijan and Canada;
• I don't have a great read on Vietnam or France as the race conditions were crazy, but I suspect I'll be very competitive with AI at my base level;
• I hate that obsolete tin can that somehow poses as an F1 circuit.  It doesn't matter what AI setting I use as inevitably I'll bust the front wing and spend the rest of the race stuck behind a Williams.

If you were to ask me, without reference to AI pace, which tracks I feel most comfortable / consistent and which give me the most problems I'd say that 

• China and Bahrain are easily among my favorite few circuits in the game.  Zandvoort has the same sort of fast / flowing feel that I really like as Brazil and I've always done pretty well at Brazil in the game;
• I know Spain and Austria as well or better than any other circuits.  I can push the limits and run consistent times on these tracks at least as well as anywhere else;
• Australia and Canada both have multiple corners which give me fits.  I've never felt like I could put a complete lap together, much less string together multiple good laps;  
• Azerbaijan is a kind of a close quartered clumsy street circuit and I don't care for those too much (especially that other street circuit that's barely fit for go-kart racing);;

So the point is that I'm quick relative to the AI on tracks I really like and feel very competent and I'm a bit slower than the AI on tracks where I have some big doubts in my head.  That seems to be on me, not the AI.

 

[Gonira 21]

I agree with you. Small deviations on AI are to be expected, but I doubt they will be as significant as our own better pace on the tracks we love and poor pace on the tracks we hate. At least this was sort of my conclusion after my comparisons with the leaderboards.

Now I'm making another experiment as I play my season.

Just after the practice and qualifying sessions (so I'm well trained on the track already), I jump out of my career and run a short quali in grand prix mode as Norris, running six hot laps (two for each soft tyre set). Then I adjust difficulty on that track so that Sainz is as close to my best lap as possible, but not faster.

This is just out of curiosity, I'm not actually changing anything on my career difficulty based on this (I decided to play at fixed level 83 for now).

Before starting my career, I did this in China, Spain and Brazil, after running a fair bit of time trial laps.

So far my results are:

Australia: 85
Bahrein: 86
China: 84
Spain: 74
Brazil: 87

From other people's feedback, I know Spain is known to be on the balanced side when it comes to AI performance, but I can't get the track, so I'm much slower there. If I didn't know any better, I'd say AI there is unbalanced. But it's not the case.

[MilligramSmile 19]

What about AI in Singapore ??? 💩💩💩
In my team career, I took 3rd place in qualification. 🤣🤣🤣
In the race he fought for the podium. 🤣🤣🤣
AI don't want to go fast... AI difficulty 102%

[Gonira 21]

3 hours ago, MilligramSmile said:

What about AI in Singapore ??? 💩💩💩

I don't know. How does this compare to your results in other tracks?

Maybe you're right. I've watched some career mode episodes from youtubers and most of them seem to get unusually great results in Singapore, even more so than in Austria. Elmo's chart and my first comparison against leaderboards also pointed to it being the most unbalanced track this year. I haven't raced there though to see how it would go to me.