The online racing simulator
Physics issues with Rfactor, something weird with roll / roll damping?
Hi guys,

I figured and hope there might be a few LFSforum lurkers here, interested in car physics, that steer WAY clear from the RSCnet rFactor forums..

By lack of somthing much better, I am busy building a Corvette C6 for that game, hoping to make it relatively good. I'm doing it as seriously as I can, even though I lack skills in some areas. Armed with MOTEC and a few tools, I hope to make it work.

Now some have suspected dodgy weight transfer / roll behaviour in ISI physics. I've stumbled on to something that I can't figure out. If you'd like to have a look, the thread is here, but skip my first post, as I obviously made a mistake there.

http://forum.rscnet.org/showthread.php?p=3449815#post3449815

The stiffness of the rollbars needed to get a certain roll angle seems way off. The oscillations from roll are such that it seems that the dampers don't work when comparing it to pitch damping..

Some odd stuff, hopefully interesting enough for a few of the Guru's here to have a look at?

/Niels
what exactly is TLLTD ?
and about the roll centers from my understanding theyre more like 2 points on the normal to the arm on which the (virtual yadda yadda) forces on the cog act

and about those oscillations ... are you sure your 2nd moment along the roll axis is remotely correct ?
Total Lateral Load Transfer Distribution :S The RSCnet thread is about a program that shows and explains that, I had not heard of it in my life.. If above 50%, more load is transferred to the front for 'understeer' and <50% makes it rear biased 'oversteer'..

What do you mean with the 2nd moment along the roll axis? Looking at a frontal view and the moment (torque) by the force (cog) on the tyre contact points?

In a side view of the car with 50/50 weight distribution, the distance from the COG to the tyres is 1.34 meter (half the wheelbase). In a front view it is 0.77 meters (half the track). I suppose that this means the forces on the suspension from roll will be greater than those of braking/acceleration, seeing as the car pulls ~1G in both directions.. But thats about 70% difference, I doubt that explains it.
Quote from Niels Heusinkveld :Total Lateral Load Transfer Distribution :S The RSCnet thread is about a program that shows and explains that, I had not heard of it in my life.. If above 50%, more load is transferred to the front for 'understeer' and <50% makes it rear biased 'oversteer'..

hm interesting not all that important when the numbers are this far off though

Quote :What do you mean with the 2nd moment along the roll axis? Looking at a frontal view and the moment (torque) by the force (cog) on the tyre contact points?

hm i think i had something mixed up there ... anyay i meant the moment of inertia in relation to rotation around the roll axis

Quote :In a side view of the car with 50/50 weight distribution, the distance from the COG to the tyres is 1.34 meter (half the wheelbase). In a front view it is 0.77 meters (half the track). I suppose that this means the forces on the suspension from roll will be greater than those of braking/acceleration, seeing as the car pulls ~1G in both directions.. But thats about 70% difference, I doubt that explains it.

wait is your problem with lat roll or long dive/squat ?
The problem is with roll, not with pitch. Or at least I think it is a problem..

I tried a few different roll inertia values from a bit low 250 .. (what is that .. kg*m2?) to really high 750.. It didn't really change what was fundamentally happening all that much visually, just the speed of the oscillations visually.. I can MOTEC a few different inertia numbers..

I tried it in LFS and its somewhat tricky because if you're not sliding at near 90 degrees sideways, there is a weird 'yaw jerking' going on when you stand still. One time it seemed to go well, using the RB4 with a similar ish setup as my corvette, and it wasn't nervously roll oscillating at all..

I know that sometimes near standstill, physics engines aren't too trustworthy, but I encounter it when driving as well, everytime it rolls it feels a bit strange..

Now ISI has felt strange to most of ''us'', however it would be nice to get some more understanding of it
#6 - w126
When you brake in the straight line the ARBs don't generate any forces, because both wheels of the same axle have the same suspension deflection. In the other case (sideways stopping) the ARBs do work and their effect at every wheel is to make the suspension stiffer. Because the damping at every wheel is still the same, therefore the system becomes underdamped, which means longer oscillations at higher frequency.
Physics issues in rFactor? What?! How dare you!
Quote from Niels Heusinkveld :I tried it in LFS and its somewhat tricky because if you're not sliding at near 90 degrees sideways, there is a weird 'yaw jerking' going on when you stand still.

I think what you saw was the car wobbling due to the tyres flexing back and forth
Quote from deggis :Physics issues in rFactor? What?! How dare you!

Have to agree on that. Niels, why exactly do you bother investigating physics issues on an ISI engine based game? IMO it is pretty much a lost cause with too many weird quirks and fundamental flaws to be worth the effort. If it was as close as netKar or DR then I could understand it, but currently (from my point of view) it just has too many wrong ideas blindly copied from old racing physics books manifested in its core. Even inputting 'correct' data seems to yield unsatisfactory results, and from what I gather, modders sometimes have to resort to drastically unrealistic inputs to get the desired behaviour. I don't claim LFS to be perfect, but regarding physics rFactor and Co are just a joke from my experience. Bah, who cares and why am I typing this?
hahaha

Of all people though, I was up there with Tristan spouting my anger and hate towards ISI. I had the not so popular 'I hate ISI' MS Paint T shirt job at RSCnet.org, for which I'll be hated by the blinder rFactor lovers until death..

Then I got busy 'modding' (what a crap word) and even though the results where not great, it did show that at least half the crapness is due to poor input data.

Now this guy with a good Corvette model wanted a physics editor, and I thought such a car might work relatively well, plus there is some data available.

Its really not THAT bad! I've improved the tyre curves and the car is somewhat close to the real specs, and drifting it around is really almost as easy as LFS! Yes I LOATHED ISI before, but it is unfair to judge simply by the available games and mods.

Next to that its quite a learning experience messing with the physics in an attempted 'good ish' way. It is kinda frustrating with LFS, though understandable, that the physics evolve slowly. A control freak like me wants to mess around with the numbers..

a short video, even the poor replay code shows body pitch / roll.. and surely this doesn't look all that far removed from planet earth?

http://www.youtube.com/watch?v=JZDu7oXAEm8

/Niels
Still looks very floaty, but that might be caused by the lack of tyre sounds in that clip. Seems to look much better than anything else I've seen from that engine, though.
It's just unfortunate that so few of the modders actually know their stuff or at least seriously attempt to get the physics right beyond arbitrary fooling around and "I think this feels right".

The question is, did you mainly use "real" data to get that result, or did you have to mess with certain settings via trial and error till you got it where you wanted? I've heard the documentation on the physics files isn't that great either...
The documentation is great, well.. except for the fact that there is none.. The devs simply don't give a @$!#. Such a difference with LFS, where you can be sure the devs aim for realism in all departments.

I entered realish data. When 'realish' doesn't give results, I don't go deliberatly 'un-real-ish' because thats a road with no direction.. Some realish numbers used in that video:

- Weight and weight distribution
- power and torque and drag, plus friction giving accurate:
- 0..60
- 1/4 mile
- 150mph points (close enough)
- 0.98 G average on a skidpad as in the specs, instant G is about 1.1
- COG height is about 40% of the roof height, as consistently seen in a SAE document
- High slip ratio modelled; grip DOES drop off to 66% longitudinally but only at 4.5x slip; i.e. max wheelspin at a standing start type situations
- Slip peak and peak change based on a real tyre ~12..20% with load variations)
- Slip angle peak for the fronts is between 8 and 12 degrees from 0 to 800kg tyre load
- wider rears with lower side wall between 7 and 11 degrees from 0 to 800kg.
- rear tyres about 5% more grippy than fronts

Not too sure about:
- tyre load sensitivity... drops to 70% at 1600kg load, in an unknown way. Real tyre data shows about 90% load sensitivity at 5000 Newton compared to 1000Newton load, hopefully I'm somewhere near that..
- Inertias, always hard to do.. they're at about 1700 / 2100 / 425 (pitch / yaw / roll)
- Aero is simple, just a 'point' 2 meters in front of the rear axle, with the right drag coefficent.
- Using a great suspension generator tool, where you can input desired camber change and roll centre height etc, I went for 50mm high roll centre in the front and 100mm at the back, with the fronts changing camber a bit more with roll than the rears..
- Its not quite clear how to do friction; as the engine / driveline friction is entered as a torque, where it would make more sense if it would be a % of engine torque instead..
Quote from Niels Heusinkveld :
- rear tyres about 5% more grippy than fronts

Surely that's just a function of tyre load sensitivity?

Quote from Niels Heusinkveld :
- Inertias, always hard to do.. they're at about 1700 / 2100 / 425 (pitch / yaw / roll)

You can get the 9 figure intertia matrix for any car in LFS from the data outputs, not sure how that compares to that 3 figure solution though.

Quote from Niels Heusinkveld :- Aero is simple, just a 'point' 2 meters in front of the rear axle, with the right drag coefficent.

Can you set the height of the drag point? If not, does the ISI engine model weight transfer from aero drag?
"Such a difference with LFS, where you can be sure the devs aim for realism in all departments."

Yeah, especially in the Formula 1 department...*cough*
#14 - axus
frankwer, please elaborate. Even if the BF1's physics isn't finished (ie. not all aspects of physics relating to it are simulated), nothing was fudged thus far. That's what Niels is getting at.

As for the inertias, the matrix has this form:
[ I_xx, -I_xy, -I_xz ]
[ -I_yx, I_yy, -I_yz ]
[ -I_zx, -I_zy, I_zz ]

I assume rFactor just uses I_xx, I_yy and I_zz, you'll have to figure out which is which in LFS.
Interesting about that inertia matrix.. One thing I'm sure can change the 'feel' of inertias, how progressive tye tyre model is. If its harsh with its transition between grip and slip, it might well feel as if inertias are low. In LFS its all slow and controlled, but that could just as well be an 'easy' tyre model and not high inertias..

How do i get that matrix for, say, the Fz50 ?
Go to LFS, load the FZ50 with whichever setup you want, remember fuel and driver/passengers will affect the output so add as appropriate, press the key "O" and a FZ50.bin file will be created in the raf folder (I think).

This is a binary format file for which there is only one known reader, which is my setup analyser (over in unofficial add-ons). Press Ctrl-R in the analyser to bring up the reader window. At which point it should be come rather self explanatory (i.e. press open, navigate to file).

Alternatively, here's one for you:
1, -7.838212E-05, -1.029422E-04
7.795897E-05, 0.9999916, -4.104151E-03
1.032631E-04, 4.104143E-03, 0.9999916

Curiously all the figures are either very small, or very close to 1. I did a little testing and the figures are almost identical for all cars. :S

Edit: OK, talking rubbish, that's the body matrix, not the inertia matrix.

This is the inertia matrix:
2063.785, 22.01738, 2.706146
22.01738, 490.5035, 52.56979
2.706146, 52.56979, 2382.168

These figures certainly vary between cars.
Thanks Bob!
If I interpret them right, i.e. the main car inertias are 2064, 491, 2382 (pitch, roll, yaw). My corvette comes in at 1586,435,1942, measuring unsprung mass around the COG of the car, with a 3.3x1.6x0.6 meter solid block of sprung weight attached. I'm using those numbers with 5% of both sprung and unsprung inertia added, still a little lower than LFS values.

Interesting to see by all accounts!


The oscillation problems are solved, it turned out to be overly soft and undamped tyres, plus the physics engine with the car near or at 0km/h isn't going to be trustworthy. The weight transfer seems to be ok in ISI, the roll per G predicted by the suspension program is matched closely in the sim, as long as you take tyre flex into account..

Bob, you mentioned tyre load sensitivity when I said the rear tyres where a big more grippy? In ISI you can set base long. and lat. grip multipliers for the front and rear tyres. So if one thinks the rear tyres, with 40mm extra thread with are 3% more grippy laterally and 5% more longitudinally than the fronts, its easy to set. Load sensitivity is also set for fronts and rears separatly, you can do whatever you want, i.e. if I want load on the rears to make them loose that initially added grip, its possible..

At the moment the biggest advantage of LFS over rFactor is the smooth tracks in LFS. Most of the ones in rFactor are worse made to start with, plus the nasty fact of driving on what seems to be the polygons, really makes most of them ones to avoid. Barcelona comes with the game, and that is pretty smooth. Really makes the ride a lot nicer and more predictable..


Is there anyone reading this who wants to have a go at a pre release corvette? Especially if that someone has been tweaking the rFactor force feedback.. It would be nice if the FF tries to keep the wheels in the rolling direction, I'm not sure if ISI thought of that possibility with their FF code..
Sure, I'll have a go Niels. I've been revisiting GTR2 and rFactor recently to see if my opinion of them has changed in the last several months and would be interested to see how your Corvette feels in-game. Where can I download it from?
I'll have a go. I have got my FF to a point where I'm comfortable with it. I still need to fight it a bit when in a slide, but it's a lot better than stock.
I've never driven a C6 though, so I don't really have a reference...
Quote from Niels Heusinkveld :Bob, you mentioned tyre load sensitivity when I said the rear tyres where a big more grippy? In ISI you can set base long. and lat. grip multipliers for the front and rear tyres.

I thought it was pretty standard for cars to wear the same tyres front and rear (although the sizes often differ)?

Quote :So if one thinks the rear tyres, with 40mm extra thread with are 3% more grippy laterally and 5% more longitudinally than the fronts, its easy to set. Load sensitivity is also set for fronts and rears separatly, you can do whatever you want, i.e. if I want load on the rears to make them loose that initially added grip, its possible...

But surely, if tyre load sensitivity is being calculated from the average load per square unit area of the contact patch, then the extra width will have already increased the area, reduced the load, and increased grip. Or are the ISI tyre physics so not so physically based? For example, just taking the load on the tyre for load sensitivity, without considering the contact patch?
ah.. load sensitivity is based on load alone, there is just a contact 'point' in rFactor as far as I'm aware. This point can be made to have more or less grip with load changes, temperature changes, pressure changes..etc..

The two posters above, thanks! you have a PM
Hi Niels,

I will have a go at it as I drive both sims.
I currently racing in the CTRA (formally STCC) working towards my Silver licence and racing in Guru, Atle is the founder which he posted a couple of messages back. I am also racing in a league using rFactor Caterham.

DDmak
Quote from atledreier :I'll have a go. I have got my FF to a point where I'm comfortable with it. I still need to fight it a bit when in a slide, but it's a lot better than stock.
I've never driven a C6 though, so I don't really have a reference...

I found that in any ISI sim, you always have to reverse the force feedback to get it feeling right (like LFS). If you are using their 'normal' FF settings, the wheel will jerk towards the apex when in a slide, and get all loose in the wrong direction/feel when you have oversteer. It feels much better if you've got the FF reversed... there should be an option for that. -100% FF in old ISI sims too .
To me it feel just 'slow'. If I'm in a slide and have the wheel turning into the slide like I should, and feel grip coming back I need to 'unwind' the wheel in a real hurry. In LFS I let go of the wheel and it sorts itself out. In rF I need to pull the wheel the right way, and it feel like the wheel is damping this motion. Ther might be a setting for this somewhere...

Edit: Found the setting, now it feels great!
It was pretty obvious, though, so I'm kinda embarassed about it...

FFB steer damper coefficient="0.07500" // Coefficient to use for steering damper. Range: -1.0 to 1.0
FFB steer damper saturation="1.00000" // Saturation value to use for steering damper. Range: 0 - 1.0

I set the damping to 0.0000 and it felt almost like LFS, only canned..
1

FGED GREDG RDFGDR GSFDG