Currently it seems the camera simulation view volume is set to be smaller than the VRC environment, with the effect that as the robot moves around mountains come into view, changing the skyline. This makes it hard for image-based visual registration algorithms, since the skyline is such a dominant feature.
I understand that limiting the camera simulation view volume speeds up the simulation (but it shouldn't matter that much if camera simulation is happening in a separate thread). I see several options.
Just increase the view volume by increasing the range to 300.
Cut the terrains down to just what is needed in the VRC. Make the VRC tasks fit in 100x100m. This is really only a problem for VRC1. Just loop the road in a race course like layout instead of an almost straight road.
Strip mine the mountains. Have no tall mountains, so stuff far away cannot be seen, even if it is there. In West Virginia they simply cut the tops off the mountains at a fixed height. You can fix the ugly scars by enclosing the VRC terrain with giant billboards (like theater sets) with images of mountain tops texture mapped on them.
Chris
Originally posted by cga on Gazebo Answers with karma: 223 on 2013-03-17
Post score: 0