Tried it without VR. It was interesting. Went to the top of the building in the second world, but there was nothing there, except white roof and a nice view of the galaxy. Performance was about 70 fps.
simply described: there is a ground mesh in the scene, a vertical raycast goes down from the camera position (yellow arrow in the picture) and determines the distance before the next step.
But it is a bit more complex. There are internal variables for the head height (adjustable via “PG up”, “PG down”), whether you are in a car, etc…
There is a virtual head point (green Point on Camera) which simulates the next step in viewing direction. If everything is ok the step takes place.
For this there is an area (red circle in the picture) where the boundary is tested, if the next step is above the ground or not. And it is tested if the next step does not exceed a certain height/depth (~0.5m). E.g. you cannot jump from the ramp.