Ask Your Question
0

Is EKF(robot_localization) expected to converge when the input data is relative like imu+odom

asked 2021-09-16 04:25:33 -0500

rajs gravatar image

Hello, I am using the EKF node from the robot_localization package for local state estimation (odom->base_link), for which I am using the data from an IMU and wheel odometry, No matter how much I tune the Q matrix, the error estimate covariance matrix diagonal elements/variances keep on increasing slowly. Smaller the values of Q, the slower the values of P increase. Is this normal when using relative measurement data? OR should the EKF P matrix converge even when the input data is relative, considering the Q and R and tuned properly.

odom0_config: [false,  false,  false,
               false, false, false,
               true, true, false,
               false, false, true,
               false, false, false]


imu0_config: [false, false, false,
              false, false,  true,
              false, false, false,
              false, false,  true,
              true,  false, false]

Thank You

edit retag flag offensive close merge delete

Comments

You can at least configure the node to stop growing the variances when stopped with

dynamic_process_noise_covariance: false
Humpelstilzchen gravatar image Humpelstilzchen  ( 2021-09-17 03:54:50 -0500 )edit

1 Answer

Sort by ยป oldest newest most voted
0

answered 2021-09-16 15:49:03 -0500

Mike Scheutzow gravatar image

No, it is not expected to converge. If this is the only source of localization data, the odom->base_link uncertainty will increase forever as time passes. There are different strategies on how to fix this, depending on what info is available.

It is common for robots with a good lidar to correct this odom estimate every few seconds. A lidar-based calculation is used to estimate map->base_link, then map->odom->base_link is corrected, and the odom->base_link covariance is reset. This prevents the odom uncertainty from ever getting very large.

edit flag offensive delete link more

Your Answer

Please start posting anonymously - your entry will be published after you log in or create a new account.

Add Answer

Question Tools

Stats

Asked: 2021-09-16 03:43:52 -0500

Seen: 20 times

Last updated: Sep 16