Why does Voxel Layer in nav2 params use scan and not pointcloud?

Question

In the nav2 params file the current setup for local costmap is using voxel layer with scan and not pointcloud, but I've seen a post where it says that voxel layer uses 3D points and obstacle layer is 2D.

ref post: what difference between voxel layer and obstacle layer?

A comment in another post says there is an issue with discrete raycasting algorithm that the obstacle layer and voxel layer use under the hood.

https://github.com/ros-navigation/navigation2/issues/545#issuecomment-459902243

The reason why I i address this is I struggle with getting points in the global & local costmap that wont decay or be removed before driving around for a while.

So does someone know why the params at nav2 uses scan and not pointcloud for voxel layer in local costmap config?

Answer 1

That's generally true, I have setup with the voxel layer for two reasons

• If your robot considers RPY using the voxel layer for local costmaps can be useful to account for the voxels while the robot is tilted on non-flat surfaces. I generally care that the costmap that the local trajectory planner is operating upon is as close to reality as possible since that is what makes the final go-no-go call.
• Secondarily, for demonstration. I've found that without demonstrations of the layers, some don't read the docs to realize there are multiple ways to configure layers for different situations and other plugin options you can use. This is supposed to spark a conversation in a team about what's right for their needs.

For a standard 2D AMR case with a 2D lidar (+/- depth cameras), I think the obstacle layer is perfectly suitable for both.

The reason why I i address this is I struggle with getting points in the global & local costmap that wont decay or be removed before driving around for a while.

Not sure what you mean by this - in effect they should be approximately the same.