Problem in fusing Stereo odometry and IMU data using ekf_localization package

[thinesh]

Hi
Recently I have been working with an AMR with Realsense d435 camera ,IMU and RTABMAP.
Since the visual odometry is getting lost while moving faster or low visual features, I tried fusing IMU data with ekf_localization.

 Mapping:

The mapping session worked explicitly with the filtered odometry.Loop closure detection also worked fine and everything was perfect.

 Localization:

When the robot is running in localization mode, it does a good job in finding the initial position.

 Localization and  Navigation

Navigation with only visual odometry

works out of the box !! No issues with that.

Naigation with visual odometry and imu :

 

Works good only in particular locations of map which is far from the start postion,but most of the time the base_link is auto correcting it's position more often which really disturbs the navigation of the robot. Like it's not able to reach the goal position since the odometry is slightly jumping around correcting itself.Especially near the start position of the robot.

 Please suggest a possible solution to overcome this minor issue

Also,I would like to know if I can use Visual odometry only for determining the initial position.After which I can use wheel odometry and IMU for localizing.

[matlabbe]

If you are navigating in map frame, the robot can jump. See definition of map frame here REP-105. What you want to do is to control the robot in odom frame instead, then adjust the goal in odom frame if there is a re-localization.

Visual odometry doesn't do any localization, rtabmap does. You can feed wheel odometry/IMU fusion directly to rtabmap node instead of using visual odometry.

cheers,
Mathieu

[thinesh]

Thanks for your valuable reply Mathieu :)