I had to resort to Blender modeled board signs, because creating ones runtime with the ShapeFactory and texturing them didn't work (see #4). If I cannot solve #4 then the tessellation should be decreased, so rendering the object would not pose as much performance impact onto the AR viewer device.

https://github.com/gdgfresno/DevfestARMap/tree/pointerarrow branch already should work (computing the direction between the world position of the target vs the world position of the scene camera - a.k.a the viewer - and turning the arrow towards that, but the arrow is misplaced in the AR scene.

Another approach to make the arrow's parent node the scene camera. In that case the arrow sticks to the viewer, but I'd like the arrow to be 0.5 m below the viewer and 0.5 m away towards the target as well.

Because of that the latest unrelated Government policy update didn't go through because Google couldn't test the app, because it is crashing right after start.
BWSSARMap_crash.txt

It turns out that Motorola Thinkphone doesn't support the Depth API (https://developers.google.com/ar/devices#google_play_devices). The app runs but never locks in the AR scene. There should be a warning.

The question is if texturing with a large texture map or rendering the numbers as 3D carved objects (Blender modeled) are more efficient. So far the textured models didn't appear on the scene at all. Try parametrize the texturing and the sampler more. Kotlin:

val texture: CompletableFuture<Texture> = Texture.builder()
    .setSource(context, Uri.parse(uri))
    .setUsage(Texture.Usage.DATA)
    .setSampler(
        Texture.Sampler.builder()
            .setMagFilter(Sampler.MagFilter.LINEAR)
            .setMinFilter(Sampler.MinFilter.LINEAR_MIPMAP_LINEAR)
            .build()
    ).build()

WrapMode.CLAMP_TO_EDGE see https://github.com/google/filament/blob/master/android/filament-android/src/main/java/com/google/android/filament/TextureSampler.java

I need to figure out how to avoid this. See screenshots later. So far I tried:

1. Disable the sunlight
   Node sunLight = scene.getSunlight();
   if (sunLight != null) {
   sunLight.setEnabled(false);
   }
2. Trying to fiddle with ambient light config.setLightEstimationMode(Config.LightEstimationMode.DISABLED);

The app would have a pre-configured off-line data for faster start-up and if the user doesn't have connection. However we could download fresher data in the background (in case we wouldn't have the latest already by timestamp) so the app could present the most current distribution fo the signs on the AR scene.

This would require internet app privileges added and the app would connect read-only to a Firebase or some exposed free, read-only data repository (GitHub or GitLab hosted?).

See:

• https://csaba.page/blog/sceneform-good-news.html
• https://github.com/ThomasGorisse/sceneform-android-sdk
• https://twitter.com/romainguy/status/1371864003882807300

If the user would log in we could connect to the conference website's Firebase back-end, pull the user's schedule and we could highlight his next session based on the current time.

This would require internet access and authentication privileges added to the app.

This might not be necessary. The app is working fine without anchor nodes as well, but the advised method is to use anchor node. Figure out if there are any downsides of the current (anchorless) approach.

Flutter has ARCore plugin ( https://github.com/giandifra/arcore_flutter_plugin/tree/master/example , https://pub.dev/packages/arcore_flutter_plugin ), but I'll need to research if all the required Sceneform features are covered.

If Flutter cannot support what we are doing then research if Unity can do it.

Extra: Terry Solis asked me: why I wouldn't consider Ionic. We have HTML5 based AR / WebVR / WebGL / Three.js solution, or more native (hybrid) Ionic with ARCore / ARKit. Article: https://www.joshmorony.com/augmented-reality-in-an-ionic-angular-pwa/