todd do you think a bicycle model with 0 long slip and very little (aka none other than little sidenotes) consideration to moment reduction from steering angle is good enough to debunk the nonsensicality of those drop off curves ?
if so ill start with my little writeup on why drop off curves are ... lets say bs
good your confirmation is all i needed
playing around with a calculator to figure out that cos is within 5% up to 18° (ie quite some amount of opposite lock) helped a bit too
Ask Pacejka. Better still, ask your colleagues Kasprzak, Milliken, Radt et al why their normalized slip angle vs lateral force curves show a drop below 90% of peak even at low slip angles as shown in this graph from the paper you cited earlier: http://i8.tinypic.com/6frvejn.jpg. Here's the link to the paper again: http://does.eng.buffalo.edu/pu ... ons/Kasprzak.SAE.2006.pdf. The issue is also discussed by Radt in some detail in chapter 14 of Milliken and Milliken and includes similar curves that also display significant drop offs below peak even at low slip angles.
I don't see how you can on the one hand criticize the slip curves in rFactor/GTR2/etc as being nonsensical for showing drop off yet on the other hand point to research which shows slip curves with similar drop offs without criticizing it too.
Support for drop off in lateral slip needs to be in the model to allow for wet conditions where generally there is quite a drop off after the peak. Very much like the rFactor/GTR2/etc curves. (Dr. Radt passed away, btw.) EDIT: Also, truck tires and so on need it.
The data posted for the truck tire was an example of drop off observed in dry conditions. This is common to see in truck tires, but I've been told several times that this is just not observed with car tires in the lateral direction by Doug and have never seen it happen myself. He was on rec.autos.simulators telling us all this back in 2002 or so, actually. Longitudinal, yes, there's often times a significant drop, and often times not.
I'm not trying to bash rFactor/GTR2/etc.. They're quite good up to the peak, so as long as you're driving on the edge and not going over it. I have rFactor and GTL and enjoy both. The original GTR press demo really blew me away and I couldn't get enough of it (the sound, oh dear God the sound!). However, it's a different animal once you go over the limit from my own sim, LFS, and Dr. Veble's model for Racing Legends. There's nothing really all that inherently wrong with the tire model I think, but they're continually giving it shoddy data at best after the peak.
In regards to my comment about 120 and 60 degree slip angles, the answer is they should be the same. The curve ought to mirror itself once you cross the 90 degree point. 170 degree slip angle is the same as -10, right?
Anyway, I assume there's some lateral scaling in slip angle with load there so they probably need the extra space in some situations, and it's pretty unlikely they ever come up with 110 deg slip angle instead of -80, so it really makes no difference either way. That comment wasn't meant to be taken too seriously. It was more of a test to see if anyone would give it some thought and provide an answer
I've done this too with rFactor and found very different results. This is totally subjective of course, but if you flatten out the curves, suddenly you can countersteer a little bit and it actually slows the rotation of the car down rather than speeding it up. To me this is all the difference in the world. My favorite non-LFS car to drive at the moment is the red Pantera in GTL. However, once I get off my own sim or LFS, I have to retrain myself to stop trying to correct a spin with a mild countersteering input because it only makes things worse. I steer into the spin to correct a slide. Again, everything is just peachy as long as I don't get too happy with the rear end
However, below or on the limit of course it would not make a difference. If you're a perfect driver you could do whatever you wanted to the curves after the peak and not notice a change since you won't operate in that area anyway. Then yes, the other parameters you pointed out completely dominate. The folks that don't particularly like ISI sims routinely complain of the "steering into the spin to correct a slide" (caused by force curves dropping off, even just a few % is enough), as well as it just being generally too difficult to catch a slide any other way. Again, the very same curves are very largely to blame for this, in my opinion.
LFS used to be the same way if you've been around it long enough to remember. There was a major tire update and the force curves were flattened out. People then immediately complained that it was too easy, but after a bit of time that quieted down and they began to notice that a lot of the unpredictable behavior was now gone. Countersteering worked, if you spun out you knew why, etc..
the curves in lfs had a rather abrupt cliff (almost as steep as the one leading to the peak) down to about 80-90% after which they went perfectly flat (afaik they have been like that ever since the first public version)
measurements done by androidxp: http://www.lfsforum.net/attach ... id=10772&d=1149022397
so once you got in the flat the cars were diftable and the steepness made the accel phase short enough for countersteering to work (polling the controller at 100hz rather than each frame helps a lot with this)
obviously steering into the spin wasnt particularly usefull either with the flatness
in a way the cars were actually easier to drift before u as nowerdays the xrt has a tad too much grip on normals for its feebly 240 hp and long links on the xrg are almost impossible now
Countersteering worked as long as you did a lot of it, of course. What I mean is using just a little bit when you've got the rear past the peaks. Once your front tires are on the right side of the peak and you countersteer only a few degrees, the force climbs upwards through some slip angle range, accelerating the spin. I remember this happening in LFS too at that time and the curves you just posted show why rather clearly. In order to correct by countersteering, you had to countersteer far enough to lower the force below what the rear tires are putting out, which is the same as you get in any sim with curves that drop off even a little bit. So once the back starts to go out a little, you need a large, very quick correction to get the car rotating back towards straightening up. Then you need to feed in the steering again very quickly as the rear tires gain grip on their way "back down" to the peak which can shoot you into a spin in the opposite direction. There's no slow, easy, back and forth steering action going on as you balance the car on the limit in harmony with the throttle. It's sort of all or nothing, and it doesn't take much of a drop off to make this happen.
Tires in the wet do this (the spinning pickup truck story), but car tires in the dry don't as you know.
so we agree after all
i was a little put off by this
which like i said isnt entirely true due to the flat
the thing with the usual isi curves is that the more you swing out the back the harder it becomes to catch it which wasnt the case in lfs
so like you said while for racing on the limit lfs was quite wrong for drifting it was more or less as correct as it is now (discounting the short transit phase which you spent little time in anyway and actually makes entering the slide easier) which explains the popularity or rather the absence of any drifters in any other sim
btw after i saw androids curves back when he first posted them i ran one of these uncatchable slides which were all too common back in p/q through one of the raf analyzers and as expected the car was sliding at ~12°
The Pacejka is outdated now, the most modern sims use a better model.
Pacejka is a magic formula, it has nothing actually to do with tyres (from a physics point af view) but the numbers it produces are very good, given good input data. Note this input data is a bunch of random numbers that makeo the model match a tyre, they are not measurements of the tyre.
More modern sims use a physically based model where the parameters of the tyre are used to model a tyre. Sounds logical but in practise is harder.
it doesnt matter which model it uses you can still measure the curves it produces and if those are off the cars will behave odd regardless of if the model employs those curves directly (pacejka) or uses some more complicated method which in the end results in those curves
I see what you mean and agree. My point was more that the LFS curves had quite a drop to them too back in the day, although yes, they flattened out at some point like you said so were not identical to rFactor/etc.. We're in agreement I think
I wouldn't say Pacejka's Magic models are outdated, really, although the tendency for sim developers "recently" has appeared to be shifting toward doing physically based ones (brush, string, etc..) If a Pacejka's Magic model does the job well enough then it doesn't really get any more outdated than F=MA. One of the F1 teams this year is using an 84 parameter Pacejka model. If you have the resources to do really extensive tire testing then it's really the way to go, although the Radt/Milliken non-dimensional model is rather choice too and expands on it.
For sim developers it's much nicer (imo) to use a physically based model given the lack of really complete sets of tire data they have access too. They're easier to work with and tend to spit out curves all on their own that are quite reasonable without having lots of crazy coefficients to deal with. It's more of a personal preference thing though, I suppose. I just like the idea of having the tire literally being simulated at the physical component level rather than having a curve fitting function, although they both can produce the same results. In the end, the curves are all that matter as Shotglass pointed out.
With 84 parameters I wonder if can still be called Pacejka... You should be able to pick some generic functions to fit all the variables.
I like the "physical" approach, too. With all those 84 parameters I don't think they can even model a flatspot, but with LFS's approach it is done quite well. And to model a different tyre you don't need loads of raw data and fit that again, but adjust some easily measurable parameters to get approximate results. This is probably the best way to model a variety of cars at the same time.
Perhaps Doug (Milliken) can comment on the data shown in figure 12 of the previously cited Kasprzak paper, which he co-authored. From that paper: "[f]igure 12 presents plots of lateral force vs. slip angle at 5 different loads and 4 different pressures. This raw data was gathered through measurement on a tire testing machine."
To emphasize the point: this is real measured data. The data shows significant drop off after peak over a range of load and pressure combinations. There is no explicit mention that truck tyres were used (though passenger cars are mentioned twice) nor is there any explicit mention that measurements were made in wet conditions. It is therefore natural to assume that the data was measured in normal dry conditions and is from normal passenger car tyres. Was the data obtained from truck tyres and/or in wet conditions without that being mentioned in the paper? I have emailed Dr Kasprzak asking him how that data was obtained, under what conditions and for details about the tyre(s) used.
Similar graphs showing significant drop off after the peak in measured data occur in Milliken and Milliken; see, for example, Figure 14.1 (which was obtained from a P195/70R14 tyre; the P indicates that it's a passenger car tyre). Was this raw data also obtained in wet conditions for some reason? Perhaps Doug can clarify that point also?
In the GTR2 curves I've looked at even at high slip angles the drop off from peak is only a few percent e.g. the curve is still at 93% of peak even at a 45° slip angle. Further, the fact that the curves extend out to extremely high slip angles (120° max) has nothing to do with their accuracy or inaccuracy throughout the range of slip angles normally encountered in-game and to characterize them by their behaviour at their extreme limits is misleading.
And I'm not trying to defend them. I don't like the way the cars feel or behave in rFactor/GTR2/etc - with a couple of notable recent exceptions in rFactor.
I haven't heard back from Kasprzak yet but, out of curiousity, I had a look as his PhD dissertation which is titled 'Extension of the Nondimensional Tire Theory to General Operating Conditions'. It deals extensively with tyre measurement and tyre data modeling and includes a lot of slip angle vs lateral force curves, some showing normalized data others showing raw data. Attached are just a handful of samples. Note that each shows a significant drop off after the peak.
Curious about where the data came from and how it was obtained, I checked the dissertation and found the following:
"Throughout this chapter and the remainder of this dissertation the development takes place with validation against two different sources of measured tire data. The first is data collected by the Formula SAE Tire Test Consortium, an all-volunteer organization co-founded and directed by the
dissertation author for the purpose of providing high quality tire data to students for use with their Formula SAE designs, student projects and academic research. The consortium is discussed extensively in Chapter 6. Formula SAE tires are small (less than 20 inch diameter) tires, normally bias ply (not radial), for use in lightweight (under 800 lb) single seat race cars. Typical tire loads are always below 300 lb.
The second set of tire data is provided by Milliken Research Associates, Inc. It is a radial racing tire used in a professional racing series from a few years ago. At the request of Milliken Research’s customer, for whom the data was initially collected, the specific identity of the tire will remain proprietary. It is a full-size racing tire with typical tire loads ranging from a few hundred pounds to over 3000 lb. Both data sets were produced at the Calspan Tire Research Facility." - Kasprzak, pp79
Given that some of the data shown in the graphs came from Milliken Research Associates and clearly shows a significant drop off after the peak, I am very curious about Todd's assertion that Doug Milliken told him several times "that [post-peak drop off] is just not observed with car tires in the lateral direction". The data clearly shows that it does.
Very interesting data there. Do you have a link to his thesis? I haven't seen it yet.
I've seen curves that drop off like this too for car tires, but only when the measurements are done on a drum tester. Once you put it on a flat belt or mobile tester (running on a real road) the curves flatten out in anything I've ever seen within 2-3%, which is pretty much within the range of noise you get throughout each revolution of the tire. Can you verify what machine it was tested on? We don't know if the loads/pressures involved are within the design operating range (talking about figure 12, not the others; maybe you can check on those?) If you pumped it up to double or triple the tire pressure and overloaded it, who knows what might happen?
The FSAE tire test consortium data is from a flat belt tester, so if you find a big drop off there then I'll have to conceed that I didn't know as much as I thought I did. They didn't run them to super high slip angles though, but at the low loads where the tires saturate at just a few degrees of slip angle, the curves are pancake flat. Edward showed rather cool videos of this. The deflections involved are stunning.
Doug will hopefully take a look at this thread. Maybe he'll have something to say to set one or both of us straight.
One thing to keep in mind with that figure 12 one is that spin can effect the lateral force considerably. In the bottom curve the cyan line has a very visible loop. As slip angle is being increased at a constant rate, the force goes up quite a bit in comparison to what you get when they return the slip angle down toward zero. Still though, the average through the two does indeed show a slight drop. Next up is to find if it was done on a drum or a flat belt. I'd bet it's a drum test...
That's the default tire and not a modded one? I don't have rFactor installed right now so can't check it, but this was over at rsc and it seemed to me this was indeed one of the default ones. Perhaps the GTR2 tires are more relaxed.
So before it even hits 45 degrees it's lost 20% of its grip already. On one real tire there was a 2-3% drop at 35 degrees (most likely within the noise threshold anyway, so I'd deem it insignificant. The transients can likely vary by more than that as pressure/temperature change during the corner), which would put this rFactor one at over 10 times the drop rate you'd get in reality. Yikes... This is what I mean by a cliff.
I know 5% drop at 15 deg doesn't really sound like much, but it's pretty noticeable in countersteering. If you had the wheel straight and all 4 tires at 15 degrees in a slide, say you had a 20:1 steering ratio, you'd more or less have to steer the wheels to a 3 degree slip angle in order to begin slowing the rotation of the car down at all. That's a 12 degree change at the front slip angles (240 degrees opposite lock! Heaven forbid you were steering into the corner at the time) in order to even begin recovering from the spin. If you steer any less than this the car's spin would just accelerate up to 5% rather than stay at a more constant yaw velocity like you'd get with flat curves. There's no room at all to do some mild countersteering when you overcook it just a touch like you'd do in a real car.
I saw videos, but haven't tried it. It does look like good fun indeed
Calspan has a flat belt tester. I'm not sure if they have a drum tester (why bother drum testing when you have a belt?) so this might be proper data after all. Looks like I might need to eat my foot here
Here's a good one that goes out quite a bit past the peaks. It's very flat and you can see how much noise is generated (2-3% variation maybe?) http://www.edccorp.com/pdfs/WP2001-4.pdf
You guys know your stuff. So could some one tell me what it is LFS dosen't do or dosent do completely that dropping the clutch @ Full boost,Redline; is always faster?
What do you mean? Shifting at redline isnt the best option. When the red light comes on its when the engine has more power in next gear. The red light isnt the redline.
Does this mean,the old stuff i learned and the feeling i got with a real car was right?A tyre will not react linear and the droppoff is always here to a certain degree?
I also got the TIR3. Its a new dimension, I would recommend it to any Sim Racer or Flight Simmer.
There is a seat out there that Simulates vibrations of the road surface, engine vibration, Grip, and so on. Its from a small company in the US, iVibe, the seat has 6 strong vibration motors who simulate all kind of things taken out of the Game engine, the effects are very customizeable and those who got it love it. I have read alot about it and I'm in the process of ordering one now, I also know somebody in my league who wouldnt Race without it anymore. I cant await to get it. Unfortunately it doesnt support LFS in a "Live" Mode, meaning it doesnt take the effects out of the Game engine but solely from the Sound, which is still good but not as good as the proper way of taking the vibrations out of the Physic engine.
As you liked Papyrus work in the past perhaps you do drive GPL and/or NR2003, because in these Sims the vibration is "Live".
Big part besides of the sense of speed is how the car is setted up. Setups in LFS are often very loose compared to normal Road cars. I'm just wondering what type of wheel and pedals are you using? There are big differences in Pedals, there are plastic ones which measure the amount of braking applied by the way the pedal traveled, or more expensive pedals that feel alot better and work by the amount of force applied to it.
You can choose to either save every Replay automatically in the Games Option Menu or do it manually by clicking "r", "1", or "s" during the Game (not sure right now which one of these
