How is Drivatar difficulty implemented


Just would like to ask how the drivatar difficulty adjustments are implemented, as in just making the cars quicker no matter what the performance of the car or is it actual driving technique.

Only reason I ask is I have found myself outpaced on long straights by slightly slower cars if you go by the performance ratings.

I play generally at highly skilled and sometimes switch to expert or pro as I find the seasons have an effect on this as well. I also use manual gears with all assists off.

I just did the Colossus race on expert and although it was Winter and I was in a McLaren Senna all the other cars were supercars or lower spec cars like the Ford Focus but tuned to the same class as my Senna (S2 I believe).

My Senna was slightly better in the S2 class as the rest of the line up, so on long straights I would think I would have an edge on most but I didn’t, in fact most were overtaking me, If you guys know the long straight at the end of this race in Edinburgh, this is where I was at top speed yet I was being catched by an Audi and a Ford Focus.

To me it seems drivatar difficulty just makes cars quicker, no matter what the performance of the car is, it should be driving technique.

drivatars “skills” are probably something like:

if player.position = 1 then
drivatar.speed = drivatar.speed * 1,2

i doubt drivatars use such useless thing like driving physics
they are sticked to the road and will do anything to stay on designed route

1 Like

Oh dont let those focus and tt’s pi foul you those cars are super op those can drive goliat around in time 8.14 or at least they are super op when ai drive those.

As previously stated the drivatars aren’t bound by the same laws of physics the player is. It is my opinion that to meet the benchmarks of difficulty they had to allow them to “break” the rules. I have no evidence of this I just try and concentrate on my own times and laps using Solo races to just grind xp and credits. This has been my experience in all forza’s all the way back to the original.

I setup a blue print on holyrood park. Ran it once on above average, I ran a best lap of 1:07 something and all the drivatars had a best lap of 1:06 or better(saw a 1:03). I won by about half a lap length.

Ran it again using higly skilled. Again I was a best lap 1:07 something. The drivatars best laps we’re 1:09 and up. Still won’t by about half a lap.

Just thought that to be a bit odd

Yeah, they’re completely fudged in the Horizon games. No matter what your PI, their cars will be scaled accordingly, but their PI number won’t really matter anyways because their performance is based on an expected race pace. Their vehicle doesn’t even matter because they’ll all perform the same whether it’s a giant truck or a hamburger box, whether it’s FWD, RWD, or AWD, and with total disregard for circumstances and road conditions.

Say your car maxes out at 153 MPH and you’re holding that speed, and you’re just barely keeping pace with the cars ahead of you, meaning they’re also doing around 153. Suddenly their straight-line speed drops off to maybe 130 and suddenly you’re catching up. Your car didn’t change. Theirs did. The game recomputed the AI pace and adjusted them down, so suddenly they no longer max out at 150 but now at 130 instead. For a while you’re having no trouble getting by them and putting them comfortably behind you, and then suddenly they’re gaining on you. Now they’re not only faster in a straight line, but they’re just as fast in a turn. Their performance was recomputed and adjusted again.

Not only is their performance completely fictional, but there’s also rubberbanding. In my personal experience, the rubberbanding seems more evident in point-to-point races than in circuit races with multiple laps, but the presence of rubberbanding has been confirmed. In another topic a user said he disproved rubberbanding, but I quickly confirmed it, and anyone here can do the same. Just go to the first race in the game (after you arrive at the festival) and do the race by cruising at a leisurely pace (I did 50 to 60 MPH). After the race, note the AI times. Redo the same event without changing anything at all, but this time race normally. After the race, compare the AI times to those from the leisurely run. The difference is huge. They won’t park and wait for you, but they’re definitely slower when you’re slower, and faster when you’re winning.

I seem to recall AR12 doing a YT video where they were testing the top speed of various high performance cars and the Senna was quite a bit slower than the rest, so I wouldn’t be surprised if you lost out on a straight.

Driving senna and losing in straight to ford focus i would laugh so hard if i didint know that is very true in that game.

And then they’d both get passed by the mighty Ford Transit Van. All praise to the Great One…

yea i feel the same there is no skills involved just AI better top speed thats why i try and tune my car for highest top speed sometimes it helps specially when u on the lower tier cars

The modded developer version of FH3 that was leaked a few years ago tells us exactly how the AI scales to the player’s PI and tunes:

% Grip Increase
% Weight Reduction
% Speed Boost

…All % relative to the player.

That’s a baseline before the race even starts. I would not be surprised if these variables were also adjusted on the fly according to how fast the player is driving during a race as well.

I think pretty much all AAA games implement AI the same way. Writing good AI is hard and takes a lot of CPU time (which consoles generally don’t have) so they’re all pretty much just play by their own simplified rule set and the difficulty setting just adjusts how much they cheat. The heavy multiplayer focus of most games now doesn’t help things at all either and doesn’t provide any incentive to improve the AI.

Hopefully the autonomous car racing scene (like Roborace) will produce some good driving algorithms that can be incorporated into future racing games instead of the cheating mess of AI we have now.