Difference in output using Odom about cartographer_ros HOT 6 CLOSED

There's no primary reference, per say. All of the data is fused together with different weights. That said, scan matching is usually the most important.

Including odometry doesn't change the scan matching behavior. Loop closure is still done and it is still done with scan matching.

It looks like your odometry fundamentally disagrees with scan matching. Check that your URDF is correct and that your odometry data is being computed/published correctly.

It's also possible that your odometry data just isn't very good in translation, rotation, or both. You can try reducing the importance of it by tweaking the sensor_bridge.odometry_xxx_variance configuration options (make them larger).

from cartographer_ros.

It looks like you might have a timing offset between your odom and your lidar.
What lidar are you using? Most have an option to sync up the the timestamps with system time.

from cartographer_ros.

Just checked your bag and config files, your Odom covariance is 0.0 for all the elements.
This will break the sensor fusion, basically making rely entirely on odom.

I would set both sensor_bridge.odometry_rotational_variance and sensor_bridge.odometry_translational_variance to something between 0.1-0.3 to start with.

from cartographer_ros.

The covariance specified in the Odometry messages are unused. Covariances should be specified using the parameters described in the ROS API.

That said, it is difficult to tune the system to handle the situation where scan matching fails. This is something we're working to improve by replacing the UKF with a separate optimization.

from cartographer_ros.

@damonkohler
I have found two observation and one prediction in pose_traker, the computing of 2 observations according to weights how to work and where is the Kraft's paper as introduced ?
I think the delta_translation and the delta_orientation should be computed differently not the same one set of algorithm. Just one idea maybe it is wrong ...

from cartographer_ros.

@simplepig I'm not sure I follow your argument. However, perhaps it's moot since we're working to replace the UKF with an optimization.

I'm closing this now as stale. Please reopen if there's still an issue.

from cartographer_ros.