Nordschleife - laserscanned?

[Simon Kaminski]

Hello, I have seen in the launcher files, that we get teh Nords.
(X:\Games\AC_TechPreview\launcher\themes\default\graphics\64\track-ring_icon.png)

Laserscanned?

Ps: I posted this in the AC Forums but it got closed. Trying my luck here now hehe.
http://www.assettocorsa.net/forum/index.php?threads/nordschleife-laserscanned.506/

Feel free to post presumptions

 

[Mr Whippy]

Yeah I was gonna say visual mesh vs physics mesh are separated these days in most cases.

I suppose if you run 1000hz physics at 36km/h then 1cm per physics step can be made.

Or at say 160km/h then it's more like 4cm... but if that 4cm distance travelled the surface goes up 1cm or down 1cm is quite important to the dampers and springs, even if it's 1cm worth of up/down noise vs being smooth it'll change the dynamics a lot.

Getting these nice surfaces is important I suppose...

Ie, if I log my car with accelerometers, at about 60mph on most roads I get loads of Z axis noise (fine bumps), but if I get on lovely new gap graded hot rolled tarmac that is also smooth (new un-opened slip-roads or motorway stretches) then you can get less than half as much Z axis noise.

Technically you'd look at both and say 'smooth' but it's surprising how much smoother some tarmac is vs others.

If that makes a difference to handling too much I'm not sure, but I can feel it in my road car, so a stiff race car driver is gonna 'feel' it I think.


I'd be interested in the process that is done on points > surfaces though. There are oodles of methods available but I wonder if end users of this data have their own algorithms/programs or just have the scan data providers 'surface' their points for them... hmmm...

Afterall, however you look at it, points > surface is gonna incur loss of information, and interpolation of information (mainly because point noise error needs to be managed too). That process is obviously very important... but as said there are loads of reconstruction methods.

If people say 1cm vertical accuracy, is that pretty much saying that a 1cm height sine shape would be flattened?


Not that it matters too much for a game, maybe they just add some noise back in for different surfaces... but it's interesting none the less.


Hmmmm...

 

[Bakkster]

Keep in mind the vertical aspect of accuracy too,

but yes, i can´t remember what iRacing´s "final" accuracy is in the game but i think they said centimeter accuracy.

From what I remember (the source has since disappeared):

Tripod scanner captures with 2mm accuracy (the mobile scanners are 4mm: 2mm for the lasers + 2mm for the IMU)
They place one guide point roughly every 10cm to anchor the splines
The splines are matched to within 1cm of the laser data at every data cloud location.

I do wonder a bit on the reasoning for using the tripods rather than a mobile scanner. If the above numbers are correct, the only difference between tripod scans and mobile (for example, iRacing Interlagos) is from <12mm error to <14mm error. Hardly significant. Makes me wonder if it's about not having to transport an entire vehicle and deal with export requirements or the initial cost of the hardware instead of the higher accuracy.

 

[Hampus Andersson]

So about 10mm difference in accuracy mobile to tripod and in-game resolution is 1cm?
In-game for AC then would be maybe 1-2cm accuracy,

Yea i wonder if mobile scanning still isn´t the cheapest way to go even though you would have to send a car across the world. The difference in scanning speed is huge.

AC being a smaller team i can´t imagine them going for the more expensive, less accurate way of doing it if it turned out to be a more expensive way to ship a car to different countries.

 

[Bakkster]

So about 10mm difference in accuracy mobile to tripod and in-game resolution is 1cm?
In-game for AC then would be maybe 1-2cm accuracy,

Read it again. Difference is 2mm.

 

[Hampus Andersson]

in-game vs in-game, i doubt it´s 2 millimeter between the two.

 

[stenyak]

Read it again. Difference is 2mm.

2mm at the exact place in the final mesh where there was a laser datapoint. Between datapoints the difference can be greater.

 

[Bakkster]

2mm at the exact place in the final mesh where there was a laser datapoint. Between datapoints the difference can be greater.

Density for either is much less than one point every 2mm. The density is measured in cm most places.

Basically, the laser in the mobile scanners like AC and F1 teams use is equally accurate as the ones iRacing uses, each point accurate within 2mm. The mobile platform adds an additional 2mm of uncertainty, but that doesn't mean the final product will necessarily be half as accurate.

We have no idea how AC is converting the point cloud into the track, but they're only starting with raw data 2mm less accurate if they're using the same type of scanners as the companies supplying F1 and WEC teams.

 

[Chronus]

Getting these nice surfaces is important I suppose...

It is really important, otherwise the difference between this method to other/traditional methods is not relevant.

Technically you'd look at both and say 'smooth' but it's surprising how much smoother some tarmac is vs others.

If that makes a difference to handling too much I'm not sure, but I can feel it in my road car, so a stiff race car driver is gonna 'feel' it I think.

[...]

If people say 1cm vertical accuracy, is that pretty much saying that a 1cm height sine shape would be flattened?

The surface in race tracks is very smooth, by comparison to the surface on temporary courses or normal roads. Even so, there is some noise generated by the existing irregularities.

I have read about methods that can be applied to noisy points clouds. A power crust algorithm (delaunay and power diagrams) can be applied to NPCs, thus improving the density. While I can't say for sure how accurate these methods are, I suspect surface reconstruction with them will allow vertical accuracy even higher than 1 cm - it also doesn't take a very long time to compute (a sample of about 210 000 points took a little over 11 minutes to compute). I see only one problem, and that is the application of these methods to a track as huge as the Nords. How long would it take to process with a fairly high degree of accuracy a segment of 4 kms?

Not that it matters too much for a game, maybe they just add some noise back in for different surfaces... but it's interesting none the less.

Why doesn't it matter? If racing sims such as NKP or AC are also expected to be of a professional grade (Stefano's presentation was also interesting in this regard), then one would expect tracks to be of a very high accuracy (as iRacing claims its tracks to be).

 

[Mr Whippy]

Some great recent additions.

I'm amazed the mobile scanners only add 2mm more inaccuracy, their inertia gear and GPS gear must be very very nice to only get 2mm inaccuracy added over the top!

That said I've been told static scanners can blow in the wind, even mounted on the usual ~ 1.5m tripods it can be a problem... and apparently with a tripod that low your range is limited to about 40m or so between scans because you want overlap for registration to avoid drift.

So they go higher with tripods, which is great, but then wind must be a big factor too... hmmm...


Chronus, I suppose the accuracy beyond a certain point doesn't matter because it's simply irrelevant because at those accuracies it's so heavily 'manipulated' or the data density is so low by these scales that you are just interpolating any way.

It's a tough one to think about in your head, but while the details are so small and the density of points so relatively low, along with noise reduction processing and the accuracy error involved, going beyond the general form of the surface is probably too much even for laser scanning right now...

Since race tracks are indeed generally smooth then that is good enough I suppose.


But I like roads too...

For now I'm getting the macro details right (ie, say a 50cm interval grid for visual), then I'm tweaking perlin noise parameters to get the 'grain' of the road feeling correct ish... (ie, details in the 0-3cm realm)

I often wonder if using texture lookups (like a normal map processed from real photos of the surface in specific lighting conditions) might be better beyond a certain point... hmmmm

Dave

 

[Bakkster]

I'm amazed the mobile scanners only add 2mm more inaccuracy, their inertia gear and GPS gear must be very very nice to only get 2mm inaccuracy added over the top!

That's the info Dave Kirkman, a Williams F1 engineer, posted on the iRacing forums on the subject. It's so low because on top of standard localized GPS compensation they use the inertial electronics from a cruise missile, which requires they get an export waiver from the US Gov't. Not sure if AC used the same grade of system.

 

[Mr Whippy]

Yeah I'm guessing each business will use their own fairly well protected IP systems to get things really good, so one mobile scanner business may be offering much different final results to another. Or if it's not something self-developed, those providing the kit will have it tightly protected.

I remember talking about inertial systems with a guy who used to develop them for submarines... obviously the kit is smaller and cheaper these days but back in the 80's they still needed accurate stuff for submarines to navigate passively... it was quite amazing how accurate they were getting back then as they had to run for a long period between 'checks' !

Dave

 

[Chronus]

That's the info Dave Kirkman, a Williams F1 engineer, posted on the iRacing forums on the subject. It's so low because on top of standard localized GPS compensation they use the inertial electronics from a cruise missile, which requires they get an export waiver from the US Gov't. Not sure if AC used the same grade of system.

Hmm, this is a bit off-topic, but I'm curious: did the US Gov. allow iRacing the use of the system? Or was this waiver requested by F1 teams?

Btw, I haven't been to the official forums in a while, maybe you can tell me, how high is iRacing's physics rate currently?

 

[Bakkster]

Hmm, this is a bit off-topic, but I'm curious: did the US Gov. allow iRacing the use of the system? Or was this waiver requested by F1 teams?

Btw, I haven't been to the official forums in a while, maybe you can tell me, how high is iRacing's physics rate currently?

iRacing doesn't do mobile scanning, the IMUs are only used in mobile scanners. Whoever uses the mobile scanner with such an IMU needs an export waiver because they are, in effect, used for weapons. Less of a waiver, more of an agreement 'I promise not to give this to anyone who will use it in a missile'.

iRacing physics rate is 360Hz, with input/output at 60Hz. The low rate is the reason they don't have fast dampers.

 

[Chronus]

Thanks, mate.

I remember reading a piece on iRacing's track scanning activities and someone saying they would or they ought to consider mobile scanning to improve the time needed to complete a scan.

Physics rate: ah, that makes sense. 4 years ago, someone close to the core team told me the physics engine tick was around 360 Hz, which is not that far away from rFactor/GTR2/Race07's 400 Hz, and is higher than NKP's 333 (Stefano or Aris, one of them posted this elsewhere, can't remember exactly which).

So...potentially, iRacing's cars, at speeds over 200 kph will not "experience" some (a lot?) of the irregularities their laser-accurate tracks are famed for (around 20 cms when ground speed is around 240 kph, if I recall correctly).

 

[stenyak]

Physics rate: ah, that makes sense. 4 years ago, someone close to the core team told me the physics engine tick was around 360 Hz, which is not that far away from rFactor/GTR2/Race07's 400 Hz, and is higher than NKP's 333 (Stefano or Aris, one of them posted this elsewhere, can't remember exactly which).

There you go: https://www.lfsforum.net/showthread.php?t=48927

 

[Knut Omdal Tveito]

Thanks, mate.

I remember reading a piece on iRacing's track scanning activities and someone saying they would or they ought to consider mobile scanning to improve the time needed to complete a scan.

Physics rate: ah, that makes sense. 4 years ago, someone close to the core team told me the physics engine tick was around 360 Hz, which is not that far away from rFactor/GTR2/Race07's 400 Hz, and is higher than NKP's 333 (Stefano or Aris, one of them posted this elsewhere, can't remember exactly which).

So...potentially, iRacing's cars, at speeds over 200 kph will not "experience" some (a lot?) of the irregularities their laser-accurate tracks are famed for (around 20 cms when ground speed is around 240 kph, if I recall correctly).

iRacing already posted that their tracks have a mesh with vertices spaced around 0.3m apart iirc. In between they use an interpolation scheme. So it's not a fundamental difference between iRacing and rFactor etc in that way. But the big difference is the accuracy of the vertex positions, not the resolution.

For racing the small wavelengths are not that important and the focus is on the ones affecting the handling (>1m).

 

[Bakkster]

So...potentially, iRacing's cars, at speeds over 200 kph will not "experience" some (a lot?) of the irregularities their laser-accurate tracks are famed for (around 20 cms when ground speed is around 240 kph, if I recall correctly).

They said in general that the actual track features are rarely smaller than 10cm.

The bigger issue is that it's too slow for separate settings for fast bump/rebound dampers. That stops stuff like the NASCAR cars where they set a really stiff slow rebound on the front to seal the splitter.

 

[Chronus]

I admit my surprise in this. Some of the comments at iracing forums hinted at a resolution a lot finer than 10 cms - 2 to 4 cms, iirc. It may not be very relevant, but considering the claims behind the quality of their laser scanning ("little ripples are noticeable at any speed" a member said once).

Fast bump/rebound: the last time I fired up NetKar I merely loaded my usual setups, so I can't remember exactly how it is with NKP. But, unless memory fails me, as I said above, NKP has a physics rate around 333 Hz. Again, I can't say for sure if this affects NKP or not, but from what I can tell and have read, it doesn't.

 

[Knut Omdal Tveito]

Yeah there's quite a few images with really detailed point clouds (not just iRacing to be fair) which kinda suggest this is what you're driving on, but it's just raw data that has to be processed into a much coarser mesh.

 

[stenyak]

Just stumbled upon this, thought you would find it interesting:
http://www.ipm.fraunhofer.de/de/pre...award-fuer-neue-laserscanner-technologie.html
(in google chrome, right click -> translate to english)