Hello Racers,
We have been working hard to get LFS ready to release. As many of you know from a recent report I have incorporated the current public tyre physics into the new development version. We call it the Retro model. It's the same tyre physics but now at 1000Hz and with a self-aligning torque component that slightly improves the force feedback. The idea is to get the new graphics out without further delaying the release to await the new tyre model that is still unfinished. We also described how Eric has expanded Kyoto massively, in a way that reminds you of Westhill but with more roads and with its own character. South City is all opened up too. Eric has continued to work on South City and Kyoto.
On my side, since the start of February, it has been mainly technical work finishing loose ends from the graphical update. In a roughly chronological order, areas covered in early February include:
Force feedback (appropriate filtering to prevent spikes)
ABS (updated for 1000Hz physics updates)
Thread-related crash related to moving subobjects
Retro model updated to use new system for friction on various surfaces
AI - some bugs were present after reinstating the Retro model
Then something I found interesting that can be illustrated with screenshots. As we now use physically based rendering and high dynamic range, the exposure (brightness of the final image) has become an issue, just as it is in real life. The difference between light areas and dark areas can be quite extreme. For example if we used an exposure level suitable for outside on a sunny day, then the inside of a tunnel or a multi-storey car park would look extremely dark, even with artificial lights switched on. The exposure must be turned up in these cases and we do that by analysing the output image for brightness and constantly updating the exposure.
So far, so good, but it's a tricky thing to get right. In a typical in-car view there is the dark car interior, the bright sky, and the landscape you actually want to see. It's not good enough to take the average of the whole scene and adjust the exposure based on that. Certain areas are of interest and they need to be exposed correctly.
Two examples that produce the wrong result if we consider the whole image.
1) In an open wheel racing car, you can see so much of the sky that the exposure is reduced and the image becomes too dark.
2) In a car with restricted view of the sky and a dark interior, a lot of the image is dark and so the exposure is turned up too high.
In the FERA, an approved mod by CarlosSainz55, the overall image is quite dark so the exposure ends up too high for the scenery.
In the FOX, the overall image is bright so the exposure ends up too low.
I used a trick, to identify the scenery using the alpha channel. When rendering the sky (each frame) and the interior of your own car in a driving view, I made sure the alpha channel of the pixels was set to zero (like transparency). The alpha channel was set to not transparent when drawing the scenery. So the image that is read to create the histogram for the exposure calculation, looks a bit like this. The magenta indicates where the alpha channel is left at zero.
Now the exposure calculation can consider only the parts that are not transparent. The exposure in driving views and trackside cameras is far more usable and stable, no longer a problem and you are mostly unaware of exposure changes as you drive around. Exposure adjustments as you move into and out of darker places seem appropriate and helpful.
One thing on my list was Z-buffer issues, that I had noticed at Kyoto and have always been around, usually seen as flickering of two nearby surfaces when it's not clear to the graphics card which pixels are nearer to the viewpoint. I noticed a post by Bokujishin in which he mentioned the "Reversed-Z" method that can produce more accurate Z-buffer results with no loss of performance. It is a now well known method in which the Z buffer values go from 0 in the distance, to 1 at the near clipping plane, instead of the other way round. I tried a quick experiment that did show an improved Z-buffer and then spent a couple of days working it in properly and solving the bugs and issues that inevitably come up when you make such a change in a complex program.
While doing that, I learned about the infinite far plane, a slight change to the projection matrix that allows us to avoid cutting off pixels that are rendered beyond a certain distance, with barely any loss of Z buffer accuracy (that had already been massively improved by the Reversed-Z system). This seemed to me almost like magic, but I tried it out and it worked perfectly. It's not the usual thing to find a little code that simplifies things and provides a better result without any downside.
One of the issues that had come up when first implementing the Reversed-Z system was fog. The haze effect that helps create a sense of depth by including more of the sky colour as objects are further away. It didn't work at all but by a simple change I was able to restore a Z value to the shaders and fix the fog. But as usual, one thing leads to another and I started to look at an ongoing problem we had, with fog glowing in dark places. The short explanation is that our graphics engine doesn't really know which parts of the air are in sun or shade, so even when the camera exposure is turned up (in a tunnel or car park) the fog effect still appears as if you are looking through lit air. And as the exposure is up by such extreme values, the fog level then appears to be incredibly bright and it looks quite bad. A previous workaround had attempted to alleviate the issue by starting the fog only after a certain distance. But it wasn't a good fix: in a long tunnel you could see a 'fog line' moving down the tunnel 120 metres in front. This can be seen in the South City Work in Progress video made by Victor in 2021 (time 2:40).
120 metres was OK for car parks but not for tunnels. In the end we came up with a solution based on a comparison between the image-based exposure, and the predicted exposure if you were outside (based on a simple calculation). Now when the image-based exposure is much higher than the predicted exposure, the fog is turned down and this seems to solve the problem in a way that it it no longer perceptible as an issue. Glowing fog in dark places is no longer, while haze in open areas is unaffected.
Continuing work after that included:
Shadows: a useful optimisation and slight improvement in accuracy
VR: post-processing is now available and final image submitted as 32-bit
VR: fix for Vive Pro 2 and any other headsets with non-square pixels
Interface: shaders to show some interface elements in greyscale
Public version: Compiled public version exe for the first time
Track editor: Some minor usability improvements
Still to do:
Support for pop-up headlights and handlebar mounted headlights
- these currently are undetected and do not cast a beam
Headlight analysis to allow smaller headlights to be drawn more brightly
- currently intensity is constant so a large headlight appears brighter
Take more steps towards building an actual public version
- currently exe runs but can only get as far as track selection screen
Some amount of adjustable weather, e.g. overcast sky
- not supporting clouds for this version but some options are possible
When will it be released?
We still can't say. There are several things on our lists and new things keep popping up, so it's not possible to give an estimate.
We have been working hard to get LFS ready to release. As many of you know from a recent report I have incorporated the current public tyre physics into the new development version. We call it the Retro model. It's the same tyre physics but now at 1000Hz and with a self-aligning torque component that slightly improves the force feedback. The idea is to get the new graphics out without further delaying the release to await the new tyre model that is still unfinished. We also described how Eric has expanded Kyoto massively, in a way that reminds you of Westhill but with more roads and with its own character. South City is all opened up too. Eric has continued to work on South City and Kyoto.
On my side, since the start of February, it has been mainly technical work finishing loose ends from the graphical update. In a roughly chronological order, areas covered in early February include:
Force feedback (appropriate filtering to prevent spikes)
ABS (updated for 1000Hz physics updates)
Thread-related crash related to moving subobjects
Retro model updated to use new system for friction on various surfaces
AI - some bugs were present after reinstating the Retro model
Then something I found interesting that can be illustrated with screenshots. As we now use physically based rendering and high dynamic range, the exposure (brightness of the final image) has become an issue, just as it is in real life. The difference between light areas and dark areas can be quite extreme. For example if we used an exposure level suitable for outside on a sunny day, then the inside of a tunnel or a multi-storey car park would look extremely dark, even with artificial lights switched on. The exposure must be turned up in these cases and we do that by analysing the output image for brightness and constantly updating the exposure.
So far, so good, but it's a tricky thing to get right. In a typical in-car view there is the dark car interior, the bright sky, and the landscape you actually want to see. It's not good enough to take the average of the whole scene and adjust the exposure based on that. Certain areas are of interest and they need to be exposed correctly.
Two examples that produce the wrong result if we consider the whole image.
1) In an open wheel racing car, you can see so much of the sky that the exposure is reduced and the image becomes too dark.
2) In a car with restricted view of the sky and a dark interior, a lot of the image is dark and so the exposure is turned up too high.
In the FERA, an approved mod by CarlosSainz55, the overall image is quite dark so the exposure ends up too high for the scenery.
In the FOX, the overall image is bright so the exposure ends up too low.
I used a trick, to identify the scenery using the alpha channel. When rendering the sky (each frame) and the interior of your own car in a driving view, I made sure the alpha channel of the pixels was set to zero (like transparency). The alpha channel was set to not transparent when drawing the scenery. So the image that is read to create the histogram for the exposure calculation, looks a bit like this. The magenta indicates where the alpha channel is left at zero.
Now the exposure calculation can consider only the parts that are not transparent. The exposure in driving views and trackside cameras is far more usable and stable, no longer a problem and you are mostly unaware of exposure changes as you drive around. Exposure adjustments as you move into and out of darker places seem appropriate and helpful.
One thing on my list was Z-buffer issues, that I had noticed at Kyoto and have always been around, usually seen as flickering of two nearby surfaces when it's not clear to the graphics card which pixels are nearer to the viewpoint. I noticed a post by Bokujishin in which he mentioned the "Reversed-Z" method that can produce more accurate Z-buffer results with no loss of performance. It is a now well known method in which the Z buffer values go from 0 in the distance, to 1 at the near clipping plane, instead of the other way round. I tried a quick experiment that did show an improved Z-buffer and then spent a couple of days working it in properly and solving the bugs and issues that inevitably come up when you make such a change in a complex program.
While doing that, I learned about the infinite far plane, a slight change to the projection matrix that allows us to avoid cutting off pixels that are rendered beyond a certain distance, with barely any loss of Z buffer accuracy (that had already been massively improved by the Reversed-Z system). This seemed to me almost like magic, but I tried it out and it worked perfectly. It's not the usual thing to find a little code that simplifies things and provides a better result without any downside.
One of the issues that had come up when first implementing the Reversed-Z system was fog. The haze effect that helps create a sense of depth by including more of the sky colour as objects are further away. It didn't work at all but by a simple change I was able to restore a Z value to the shaders and fix the fog. But as usual, one thing leads to another and I started to look at an ongoing problem we had, with fog glowing in dark places. The short explanation is that our graphics engine doesn't really know which parts of the air are in sun or shade, so even when the camera exposure is turned up (in a tunnel or car park) the fog effect still appears as if you are looking through lit air. And as the exposure is up by such extreme values, the fog level then appears to be incredibly bright and it looks quite bad. A previous workaround had attempted to alleviate the issue by starting the fog only after a certain distance. But it wasn't a good fix: in a long tunnel you could see a 'fog line' moving down the tunnel 120 metres in front. This can be seen in the South City Work in Progress video made by Victor in 2021 (time 2:40).
120 metres was OK for car parks but not for tunnels. In the end we came up with a solution based on a comparison between the image-based exposure, and the predicted exposure if you were outside (based on a simple calculation). Now when the image-based exposure is much higher than the predicted exposure, the fog is turned down and this seems to solve the problem in a way that it it no longer perceptible as an issue. Glowing fog in dark places is no longer, while haze in open areas is unaffected.
Continuing work after that included:
Shadows: a useful optimisation and slight improvement in accuracy
VR: post-processing is now available and final image submitted as 32-bit
VR: fix for Vive Pro 2 and any other headsets with non-square pixels
Interface: shaders to show some interface elements in greyscale
Public version: Compiled public version exe for the first time
Track editor: Some minor usability improvements
Still to do:
Support for pop-up headlights and handlebar mounted headlights
- these currently are undetected and do not cast a beam
Headlight analysis to allow smaller headlights to be drawn more brightly
- currently intensity is constant so a large headlight appears brighter
Take more steps towards building an actual public version
- currently exe runs but can only get as far as track selection screen
Some amount of adjustable weather, e.g. overcast sky
- not supporting clouds for this version but some options are possible
When will it be released?
We still can't say. There are several things on our lists and new things keep popping up, so it's not possible to give an estimate.
, I'm sure some of us would be more than willing to help beyond just enjoying the shiny update (although many would likely just complain about the crashes instead).
