Generic collision detection with 3D Tiles is pretty tricky. The current camera controller reads the depth value in front of it to determine if it collides, but isn’t aware of things to the side. I don’t have a good solution right now, just a complicated one where you store bounding boxes for each building and check every frame if the camera is within those bounds. There’s a lot of details to work through, so I’m not sure if I would actually recommend doing that.
For placing models such that they don’t collide with the tileset, you would need a similar approach. Store the bounding boxes of all the buildings as part of the tileset metadata (via the batch table) and reference those before placing other models.