Laser scanned tracks

You can ask the same way about physics. Who really cares as long as fun from competing against each other is good?
For someone it's enough to just race. For other ones it is important to drive in environment as close as possible to its RL equivalent.
I believe that simracers usually chose second option.

 

I remember sandbox. Actually, I'm using sandbox for building tracks. Primitive, but easy to use until you start creating multi-course tracks. It can be used to make accurate tracks. Although I do wish it had more features.

I think that the spline method is the best because it creates a flat surface to run on whereas the mesh method creates a series of planes that changes your car's orientation. This creates bump that may or may not be noticeable depending on the size of the polygons. Plus, the more polygons you put in a scene, the slower the computer runs. So I have two reasons for my opinion on why "spinlines" are better. (1) Performance, and (2) maximum accuracy.

 

I try to drive on what is real as possible. If a track is inaccurate because someone just wanted to to make it raceable...then that track should not be considered a representative of the real track. Lets use Laguna Seca for example. If someone adds grip to corners, takes some away elsewhere then that is no longer Laguna Seca and should not be called Laguna Seca.

If you try to get the track as real as possible then you can call it that, but if it doesn't represent the track in the best way possible then it shouldn't be called the same.

That is a problem I have with most rFactor modders out there who release "Track A" wanting people to think it is really "Track A" when it isn't. It is their vision on how that certain track should be.

Hopefully I put that in a way people can understand what I meant, my vocabulary doesn't kick in again until the weekend.

 

I agree. A real track should look as much like the real thing as possible. If it's a fictional track, then it should still look like it could be a real track. However, we've seen that most simulators only deal with the track geometry that is between two walls, then place objects on the infield/outfield somewhat randomly and/or have a low-resolution image of the infield. So when driving on the track and viewed from the track cameras, it looks like the real thing, but when viewed in a scene viewer, like 3D SimEd, it looks nothing like the real thing. I don't mind if at low detail, the scenery is not visible, but at max detail, it should look like the real thing even in a scene viewer.

Same with a car. And this is one area that NR2003 actually got right and other simulators skipped on. In the simulator, a car can look like the real thing, but what it comes down to is how the meshes line up on the scene viewer. A model with overlapping polygons is not what I consider a good-quality model. There are some instances where overlapping is necessary, such as on an object that rotates on an axle, but for the most part, overlapping should be avoided as much as possible.

 

I'd pay 10 bucks for a road circuit but only if I could use it offline as well as online.

 

I never understand this stuff. Are you saying that the physics engine is reacting to the laser scanned grid/coordinates (albeit a simplified version of) and that that's why we see the tyres floating over and sinking into the visible track surface that's generated by the graphics engine? Because if so, then why are there visible features e.g. the dip at T2 Lime Rock (where the grey patch is wider) that the tyres float over - why create a visible dip that isn't there and will be ignored by the physics even if you park on it? Or are you saying that the gfx engine is generating the visible suspension and wheel behaviour but that the laser grid is the track surface we see? There are also invisible bumps the suspension reacts to. Or something else?

Anyway, I would point out that if you're cornering on the limit, the smallest bump/irregularity can create tragic results. In iRacing you'll see people spinning out regularly at the same places because of these features that are actually too small to see. So, as has been pointed out, unless you get every little bump and detail right, including rescanning when surfaces are relaid, then it's never going to actually be the same as the real track for racing purposes. Which is why some pros like iRacing tracks. But then the physics doesn't necessarily accurately portray the behaviour over these details anyway. So I suppose if you get the basic undulations (and shape and number of corners ) right, then that's all that can be expected. Really that's why you pay iRacing for this level of detail (and rescanning too).

I have just quit iRacing after a year btw. I found it a depressingly sterile, unimmersive experience mostly because of the sound and toy-town graphical appearance - I never forgot I was at a PC - unlike many other sims (RBR, GT Legends, rFactor). And just trying netKarPro really convinced me with the feel that I've had enough of iRacing. Another thought: I wouldn't mind paying for laser scanned tracks if you kept them; whereas all my iRacing content is now inaccessible without a subscription.

 

It's like, 2 pieces of mesh, one is drivable and not rendered (the laser scanned bit) and the other is not drivable and is rendered, meaning it doesn't react to your car, it's just there for looks. So what you see is not what your driving on.

 

I think that using this method is the best way to go. Since one mesh is extremely high resolution, it would be the best thing to drive on, but if it was rendered, it would crash the system. The other mesh has a variable resolution so that the simulator doesn't use as many system resources when rendering the track, but if you drove on it, you would crash the race car.

 

I hated watching replays in iracing and watching parts of the car dip into the road even though the car wasn't actually hitting it.

 

Thanks for the explanation. It does still seem odd to me though that there's a large dip rendered visible that the external replay of the car (which seems to agree with the cockpit view you drive with) floats over, i.e. it seems it's rendering a large feature that isn't there in the laser mesh you drive on. I'd be expecting the tyres to be momentarily rising above and dipping below the 'simplified' rendered surface rather than the other way around.

Using the camera tools to bring the eye line down to track level, you can see the size of the discrepancy between what's rendered for the track and where the tyres are shown to be. I don't understand why cars need to float with up to 4 wheels considerably away from the rendered surface for the entire length of flat (judging by cockpit response) straights. This is not to do with netcode, BTW, as it's happening in 'offline' test mode.

 

Interesting blurb on scanned tracks by a simbin developer.

"Hey Guys,

Nice discussion you got going.

I can only share my own views on the matter not the SimBin view but i would like to mix in on the discussion a little bit.

First off all there is cost involved that are considered to be great for a company as ours. In the end it has to be paid by someone somewhere. And development budgets of our size do not have the room for laser scanning at the moment. That is the boring bit.

In theory in my research, and its just theory until we have been able to try it ourselves. I see another "issue" with track scanning. And that is that even with track scanning that allow you to scan up to MM accuracy the current hardware does not allow you to use 100% of the data collected.

For example, a single bump in the track in the cloud data consists of 1000 points (number is hypothetical), that does not mean we can use a huge amount of polies/triangles to make that bump accuratly visible on the track it sef. We would explode our poly budget per track if we want to take full advantage of the cloud data.
So we would be investing a big amount of money to gain that enormous amount of additional data, but that data we can only use to interprete the track surfaces with an X amount of polys. Basicly ignoring 80-90 percent of the data to make it work, therefor wasting the actual accuracy you gain with track scanning.

So the deicission to spend a (big) part of a development budget on laser scanning and then ignoring the time, money, effort and data resulted from that is a hard sell to possible investors and publishers.

Having said all of the above, i might be completely wrong. I hope the day will come where we get to try this method and see and feel its results in our production pipeline. And then maybe who knows.......until then we will do things old school, with track visits by the reference team, receiving CAD data, and shoot thousands of photos on track to still maintain a high standard of track modeling."

 

Yes. He is right partially. And it is a proof how to manipulate a numbers to reach a point

It is true that with today computers we cannot utilize whole point cloud in real time simulation. But we can reproduce surface variations with some defined precision. And this accuracy is The Thing I'm talking about. It is impossible to get bumps using blue prints or other CAD data, because such bumps are not mapped into such documents. Usually there are flat surface maybe with new added patches. There is no bumps caused by erosion or surface waves created while making track or caused by temperature.

Additionally there must be inaccuracies in blueprints comparing to reality. Is quite normal, that track is not build with 1mm accuracy (I mean shape) from documentation. If blueprints are made based on existing track, it will be inaccurate also, until are created on laser scanned (or other reproduction techniques) data.
As stated a few times already, SBT's Spa is meters way off comparing to real shape (and I'm not talking about bumps!). So maybe they should consider to change data source.

Finally I can bet, that laser scanning (including processing data) is faster method than measuring objects, drawing objects and positioning it with mm accuracy.

Of course using already prepared data will be cheaper than scanning it by own. Maybe there is a free space for companies (maybe like google) which will be own great database of scanned areas.

 

That's where iRacing does the laser scanning thing cleverly. They do NOT show each bump nor represent it as polygons, still all the information is there for the physics engine. They have two separate layers. One is the "physics" layer, an underlying mesh, which includes full data down to millimeter resolution and the other is the graphics layer, the one the player sees. This is why the iRacing cars can look a bit weird in replays and be seen "dipping" under the surface of the asphalt.

This is also partly the reason why the iRacing force feedback is way superior to almost anything else.. it doesn't hit a "polygon edge" limit. It has much more data to work from, thus being smoother on crests, hills and bumps (which need much more polygons to be smooth).

So, in reply to the Simbin guys: You need to think outside the box a bit. Everything does not have to be represented by polygons!

Cheers!

 

@bmanic. Are you an iracing track developer?

 

If it would be up to me, I would represent bumps as heightmap. You can calculate normals from heightmap on the fly. Milimeter precision as mentioned above is still too much for today's computers, so I don't believe iRacing has it. Bump resolution of 1cm data density is possible however, and is really good enough. Besides, iRacing doesn't simulate tyre at 1mm precision (and I think even NTM will not provide that), so talking about 1mm precision is just dreaming.

I'd say that technically it's no problem to describe racetrack at high precision. The problem is always obtaining and processing real-world data and then developing a simulation engine that really makes use of this kind of data.

 

Hell no!

It's pretty common knowledge. It's been discussed on their blog several times and in the iRacing forums over the years.

Cheers!
bManic

 

I think in iRacing they choose which bumps to model accurately and which not. They have all the data they need and can update the tracks in the future if/when there are more powerful computers. A bump at a critical point (like the damn bumpy T1 in sebring) of a track is more relevant than some bumps off the racing line.

This is just a guess but there is no denying that the bumps are VERY accurate in iRacing. Try driving very slowly over some bumpy parts of a track and you'll quickly notice how different each bump within a cluster feels/looks on the replay and force feedback. It's pretty amazing in my opinion.

Then again, all this extreme detail is a bit silly in the long run because the real life tracks change all the time. If a corner is too bumpy and in dangerously bad condition the track is repaired! Suddenly the data in iRacing is pretty useless (unless they go back and rescan the track surface which I highly doubt they will). A lot of people think that iRacing's ultimate selling point is the tracks.. well I disagree. There are plenty of extraordinary tracks for rFactor/SimBin titles. For me it is the community and the almost certain guarantee that the races are relatively clean and civil (this is of course not always the case).

Cheers!
bManic

 

Ah o.k. Just sounded like you were doing an epidsode of iracingtv there. lol. Yeah so the mm accuracy thing is very misleading at best.

 

Just wanted to note that (at this point) there won't be any laser scanned tracks in the initial release of rFactor 2. Although (like with rFactor 1) there's nothing to stop anybody doing what you are suggesting. I know how expensive the process can be.

What I liked about laser scanned tracks was the fact that the surface felt 'alive'. I honestly didn't mind about the realism (I was never going to drive on these tracks) but the track surface not being totally smooth made a big difference. For me, this feeling is already there with rFactor 2 because of the combination of weather, real road (track surface) and tire models. These make the surface you're driving on feel very much alive. Additionally there are big steps made in the bump maps.

Points made about track changes are quite valid, a remodeled corner (or entire track) is a big problem. While a 'perfect' track could be made, you can't really go from sim to real instantly anyway, because of the small physics issues you would obviously encounter by doing so. For example with stock cars, they change the tires they use every darn week, this changes corner speeds and handling of the car so you're going to have to adapt your driving anyway even if the physics model was absolutely perfect (which really isn't possible)... So a basically realistic track (made by any method - it just takes hard work) might actually be as much useful as a laser scanned one. You're still going to have to adapt yourself to drive the real thing.