visual odometry drifting

[osmancns]

Hello...
I connected output of rgbd_odometry (with <remap from="odom" to="vo"/>) to robot_pose_ekf.
And ı have a laser and ı publish odom message with laser_scan_matcher ( I convert pose to odom ).

if I use robot_pose_ekf with only odom topic ( from laser_scan_matcher), no problem everything is good. But
ı want more stable odometry.

if I use odom topic with vo topic ( from rgbd odometry), when my robot started to turn, my odometry drifting increasingly. ( odometry İS providing for AMCL)

why odometry are not more stable with rgbd odometry ?
what is problem ?

thank you...

[matlabbe]

Hi,

If odometries of each source are "ok" individually and that merging them into robot_pose_ekf produces bigger drift, it may be a problem with the covariance values contained in odometry messages (saying that, but I am not an expert of what kind of covariance robot_pose_ekf needs). Make sure that you don't have multiple nodes publishing the same TF (by default rgbd_odometry publishes /odom -> base_link, to disable set publish_tf=false)

Note: you didn't mention rtabmap node, but if it is the case, you cannot use AMCL and rtabmap node at the same time.

cheers,
Mathieu