End goal here is to get ~same accuracy as what ARKit's vio provides, but its unclear if I'll be able to reach the same on device speed. Thats the stretch goal =]