camera.flyTo goes underground

[angrycat9000]

camera.flyTo ends up underground in a location where the globe does not have a height.

The expected result is that that camera flys to 100 meters above the ground at the given location and the vehicle is visible on the surface of the ground.

Sandcastle example

Browser: Chrome 103.0.5060.114

Operating System: MacOS

[mramato]

Regarding:

let terrainHeight = viewer.scene.globe.getHeight(cartographic);
console.log(`terrainHeight=${terrainHeight}`);

It's correct behavior that terrainHeight is undefined here because globe.getHeight only works with terrain tiles that are actually loaded and only suitable for querying data as-is on-screen. You need to use the async sampleTerrainMostDetailed to get the true height of terrain at a location, which will definitely return a height.

I still think there is a bug here because flyTo should not end up underground, but you are telling the camera to fly underground so we should clarify exactly what the official behavior is supposed to be. I expect the camera terrain collission to handle this though and pop you up above grown at that location.

If the goal is to fly to the entity, you should be using viewer.flyTo(entity) but the biggest issue here is that viewer.flyTo(entity) doesn't appear to work either when heightReference is CLAMP_TO_GROUND.

The below tweaked code demonstrates what is going on here. Click FlyToModel and it will be incorrect. Click it a second time, and it will be differently incorrect. Click it a third time and it is correct. There is clearly a race condition going on involved with clamping to ground.

const viewer = new Cesium.Viewer("cesiumContainer", {
  terrainProvider: Cesium.createWorldTerrain(),
});

const longitudeDegrees = 149.515332;
const latitudeDegrees = -34.984799;

const position = Cesium.Cartesian3.fromDegrees(
  longitudeDegrees,
  latitudeDegrees,
  0
);

const entity = viewer.entities.add({
  position: position,
  model: {
    uri: "../SampleData/models/GroundVehicle/GroundVehicle.glb",
    heightReference: Cesium.HeightReference.CLAMP_TO_GROUND,
  },
});

Sandcastle.addToolbarButton('FlyTo Model', ()=> viewer.flyTo(entity));

[javagl]

^{I'm occasionally thinking "Well, athough I'm not familar with CesiumJS, I'll have a look at this bug - it cannot be so hard..." (and I'm regularly proven wrong...). But even at the risk of distracting others from the actual cause here, two observations:}

When enabling
Cesium.ExperimentalFeatures.enableModelExperimental = true;
in the last sandcastle, then clicking the "FlyTo Model" button does reliably zoom to the model:

But... the model itself is underground.

I started diving into the rabbit hole of comparing Model and ModelExperimental, and noticed that ModelVisualizer.prototype.getBoundingSphere computes significantly different bounding spheres...

Even though the model has the same modelMatrix in both cases. I haven't been able to figure out the reason for that one, but maybe someone can narrow down the search based on that...

[lilleyse]

@javagl are you on a commit pre-#10395? @j9liu might have more background here.

[javagl]

I initially compared the behavior of the public/hosted 1.95 with the local main, but then noticed that the behavior of zooming to the model (but the model itself being placed incorrecly) was also happening with enableModelExperimental.
(I even went back to 3d138ad , wondering whether this might be related, but at that point, I was just poking around in the code: Iff this was related, it would probably have been noticed earlier...)

[j9liu]

I'll look into this later today. It's strange that ModelExperimental doesn't seem to be accounting for heightReference at all; the model stays underground the entire time.

[mramato]

I labeled this next release, since ModelExperimental seems to add an additional regression on top of the existing problem.

[j9liu]

I confirmed that heightReference works when directly passed into ModelExperimental via #10448 (local Sandcastle for proof). It works for the "Clamp to Terrain" sandcastle but changing viewer.trackedEntity = e to viewer.flyTo(e) reproduces the buggy behavior. I'll have to see if there's something in the entity API that ModelExperimental isn't interfacing with correctly

EDIT: I did find a typo in ModelExperimental: one of the functions modifies _heightChanged instead of _heightDirty. Still doesn't solve the error, though, I'll keep digging.

[ggetz]

I labeled this next release, since ModelExperimental seems to add an additional regression on top of the existing problem.

@lilleyse Does it make sense to address both the model experimental "regression" and the underlying problem at the same time, or are these two distinct problems?