DevLog #1

Lately I've been exploring asset packs and realizing that there's whole worlds out there that have yet to be seen in VR.

When I saw this post on reddit I realized there really wasn't a lot of cyberpunk content on Quest, and thought maybe it would be fun to make a little demo out of what I could find on the asset store.  I did a search and found a lot of really good stuff by IL.ranch and had OP pick something in my (meager) price range.

I really was expecting more of a fixer-upper at the price point but no.  Great asset with a great demo level that already feels alive.

Initially, I imported the project into my BaseVRProject and ran it in editor with my VR character controller.  It looked great and worked very well.  I did notice that there was not robust collision; only where the player was expected to be.  This is good but not appropriate for how I'd like to be able to explore so I had to disable collision checking for my climbing movement system; this means you can actually climb in air.  It's good for debug but bad for immersion, generally.  One of the first things I'd like to fix going forward will be the collision issues.

So now I had to see how this ran on Quest.  Keep in mind, the initial asset import took my computer hours to crunch on, and the initial build for Android was similarly long.  And not without issues either.  I had to make adjustments to several shaders (and remove a few) just to get everything to build.

The initial build had major graphical issues when I'd look in certain directions from certain positions.  They seemed to be related to performance (though a later test had me doubting that) but I had seen similar problems before with Single Pass rendering so I switched it over to (Leeloo Dallas) Multipass.  Also the cull point and LOD systems were way too aggressive so I increased the LOD bias to 6(!)

And it worked!  Slower of course... and it was already pretty slow.  Here's a video of the build in action.

Subsequent builds and tests had me changing traffic parameters, making the moving cars and other more minor textures use SimleLit rather than Lit materials and improving mipmapping parameters.  I also baked occlusion data which was a big win.  Managed to gain a lot of framerate without losing much visual detail, but it's still not up to 72.  I did have a build that ran locked at 72 which forced LOD1 but the LOD1s are not detailed enough up close to make it a viable option.

Ideas for further framerate improvements: adding a new LOD that switches between Lit and SimpleLit.  And use Simple lit on LOD1 (would now be LOD2).  I'm still using all real-time shadows too, because when I started baking the lightmap it said it would take 20 hours... and it basically locked up my computer during the process so I'll have to do that another time.