Understanding Covariance matrix role in Kalman Filter in robot_localization
I am using robot_localization package for state estimation. I am fusing Odometry data and a sensor data with providing values to civariance matrix of each message. Now i have few confusions regarding how things work.
My sensor provides x,y positioning information with a fixed error of 0.2 m for every position,The error is fixed. And the odometry error grows over time/movement. Should i provide a fixed covariance matrix values to both or how can i tell the system that odoetry error is growing?
If i use velocity from odometry instead of position for state estimation, then what will be the difference in selecting the covariance matrix.
For my sensor covariance should be 0.2 fixed, right?
Any suggestions on how to calculate growing odometry error covariance matrix instead of fixed? I have measured the odometry error variance and standard deviation, should i use that or something else?
Asked by enthusiast.australia on 2019-11-12 10:29:08 UTC
Answers
For 1, 2, and 4, just compute the velocity covariance for odometry and feed that to the filter. The EKF will take care of the rest.
For 3, see https://answers.ros.org/question/334901/covariance-matrix-for-custom-odomtery-messages/
Asked by Tom Moore on 2020-01-06 05:37:20 UTC
Comments