[ros-users] costmap_2d elevation questions

[Jack O'Quin]

It may seem nonsensical to ask about the "height" of a 2D costmap, but
I am confused by what we saw today and could use some help making
sense of it and deciding what to do.

As mentioned in an earlier message, I am experimenting with
costmap_2d_node to see if our autonomous vehicle can use it for sensor
fusion. It seems to work OK in stage simulation. We are running ROS
C-turtle on Ubuntu Karmic and Lucid.

Today, we experimented with it on the actual car feeding it an
obstacle point cloud from our Velodyne 3D LIDAR. I can view the
PointCloud in rviz and it looks reasonable. But, when I view the
obstacle GridCells, none appear. However, rviz can display the unknown
GridCells (which is all of them). Strangely, they appear a couple of
hundred meters below the car and the obstacles. This offset looks
close to the actual elevation of our test area (209 meters above sea
level), so I am guessing the map may think it's at elevation z=0.

Our GPS odometry source publishes the actual elevation for the
nav_msgs/Odometry message and /odom->/vehicle transform. There is
(currently) a static identity transform from /map to /odom (0,0,0
translation and 0,0,0,1 rotation quaternion). So, I expected the map
to appear at the same elevation as the vehicle and the odometry. I
can't be certain whether this is purely an rviz display issue or
inherent to the costmap, but I suspect the latter, because it would
explain why no obstacles appear in the map, they are all way above the
/costmap/max_obstacle_height parameter setting (2.7m). I did try
adding 209 to the max and min obstacle heights to see if that would
make a difference, and it did not.

Another possible explanation is that the velodyne points are published
as "/velodyne/obstacles". I can't tell whether costmap_2d allows a
<sensor_name> containing slash or not. Is it legal to set the
/costmap/observation_sources parameter to "velodyne/obstacles" and
then define parameters with names like
"/costmap/velodyne/obstacles/sensor_frame"?

Any ideas what is going on and suggestions for what to do about it?
-- 
 joq