I was having little fun race with AI, only 10 DTM92 cars at Nordy, AI strength was set to 88 which made them bit too slow. Funny part was that I was at 1st place with Mercedes, 1 minute and 26 seconds behind me was Kurt Thiim AI, when I was about half way trough that longest straight, traveling over 250kph, I did see Kurt Thiim was gaining at me, he gained one second during 1/4 length of that long straight! I know legends were fast, but that made me wonder why AI gained so much on such place and while I was on some other section of track spinning 360 and hitting wall, AI still was slower and gap got bigger. So this leads to question how AI times come to be, there must be some kind of mathematic formula I guess, it can't be that they would drive like that, at my speed over 250kph Kurt Thiim would of needed probably well over 300kph to gain so much on such short distance? Has anyone else noticed some funny time gains by AI? Oh and my times surely were not the most awesome, around 7:24, from some reason losing few car parts is not helping with the laptimes, even one would think car becomes lighter
This is more about how the gap is calculated rather than the ais lap times. But i asked about gap calculation a while back and dont think anyone could tell me how its done so i cant help you here either
From what I have read on this forum, AI uses the same cars (physics model) as the player. So they can´t go 300km/h in a DTM92 car. AI therefore are slowed by lower cornerspeeds, early braking points and wrong racing lines I have observed
I'd guess the gap calculation is done by calculating the time gaps at set points on the track, let's say every 50 metres. So the gap you see is continuously being calculated on what you have already done. On a short circuit it will look normal but on the nords you think you should be pulling out the gap as you are in the straight but actually the gap is calculated on what you did earlier. So thimm was going through the earlier section quicker than you did and thus the gap is falling. Think of it like sector gaps, but there are hundreds of sectors instead of 3.
Thats a way of calculating it that id find quite reasonable. But i dont think thats how its done in r3e. Cuz then youd be say at the exit of the small caroussel (whatever way you spell it ) 15 seconds later than the leader and the gap youd be shown is 15 seconds. But then arriving on döttinger höhe the gap goes down to say 10 seconds without me being 5 seconds quicker through the one corner in between those two sections
Interesting. I wonder if there's a layer of prediction going on as well then? I've never really noticed the issue but I don't race the nords much..... Seems to work on on short tracks.
Afaik the delta calculation does indeed use subsectors to calculate the lap distance. There's no prediction going on that I'm aware of. It will always be a bit rough on Nordschleife simply because the track is so long and you can't have unlimited subsectors.
To be honest I don't use the game delta anymore, I have second monitor app which displays lap pace against opponents. It turns green if you are faster and red if you are slower, gives a good indication of whether you are catching the guy in front etc. I don't like on screen data at all.
Just fyi, you can also get the delta on all in-car data displays, on some you have to switch pages for that tho.
But surely the gap only gets updated when youve also crossed the line of a subsector right? And in that case thered still be a new subsector every 5 metres or even more often which doesnt sound like several seconds should go missing from one to the next Another question that came up regarding your non prediction thesis is how it works when somebody spins. Say youre 10 seconds behind another driver who then spins. If the gap is just based on how much earlier he was wherever i am now then the gap would stay around 10 seconds and then suddenly drop to 0 once i get to him. But if someone in front of me spins i can see the gap getting smaller and smaller until i pass him instead of the other scenario. How does that work without any prediction then?
I have a little dash screen that blocks my view of the in car dash. I have lap delta etc on there so it's all good. Plus, depending on the car the in car dash is not always visible.
I don't know the actual figures, but I think for the in-game data displays the subsector distance on Nords is considerably larger than what you assumed (in other words, there's fewer subsectors). There would have to be 4000-5000 subsectors in your scenario, which I think is way too many. What I do know is that the webHUD produces a more accurate delta because it uses a finer set of subsectors. Regarding the prediction question, I think I may have misinterpreted what you guys meant by "prediction". Of course there has to be an approximation in between subsectors. Otherwise, as you already pointed out, the delta couldn't change before you reach the next subsector. I don't know how exactly it's done here, but I imagine the subsectors are equally spaced and with your current speed you can then calculate an approximation of when you're going to reach the next subsector.
That would be the first sim where this is the case. The strain on the CPU would be overkill. Do have a link?
Im always talking about the live gap to your competitors that you can see on the position bar which is basically constantly changing hence my guessing very short subsectors. Just to make sure were not talking about different things That would then explain the sudden gap drop when entering döttinger and is what i too think is how r3e likely does it
Only game I know where AI is actually driving the car like player and using exact same car is BeamNG.Drive and there AI is really heavy WIP currently. AI knows where road is as roads on map have that information, but AI is calculating braking, speed etc. like pretending to be a human driver. That is not heaviest part to compute though, it's more of physics sim (softbody physics, 2000 times a second chassis flex, stress etc.) so not much of sims comes to close how heavy thing is to run. Difficulty to pull of working AI with that approach is something insanely hard in that game compared to what most racing sims do (and in current WIP state AI does not really race, but makes wonderful DUI simulation). In rF1 I think waypoints gave speed targets to AI, but AI cars used different physics, it is interesting that Raceroom has same cars for AI. This changing gap I did observe, now on top of screen when you are racing there is constantly updating gap to all cars behind you (or at front of you) it looks to be realtime, so I guess it is somehow smoothed value between those update points so it looks realtime. As I understand what was explained to me, from update point to update point time might be gained, but it is not instantly snapping to new value, but instead it gets some form of smoothing so gap varies a bit as it takes time to adjust to new value smoothly. Edit: I did made futile attempt to make my text more readable by adding more words.
Addition to my last post If youre gonna make that distance/speed calculation anyways you could just as well constantly calculate the distance to the cars around you, take the current speeds and calculate how long itd take everybody to drive a certain distance at their current speeds and throw out the subsectors completely. Forgot what i wanted to get to with this but maybe yall can figure it out
I use distance/speed calculation in my app, when sim does not provide gap information. And after a certain distance between cars it becomes really inaccurate, because it ignores the profile of the track. If you're on a straight, going fast, and the car in front of you is in a slow part of the track, using your speed for calculation basically assumes you would be able to drive the slow section at unreasonable speed. On the opposite, using the opponents speed basically asumes you will be crawling on the straight. The only benefit is that it is cheap and easy Which why I use it