Update odometry based on pose estimate

edit

You do not need to worry about the drift in the odometry coordinates unless you start seeing massive jumps between successive readings. Gradual drift over time is fine.

Odometry typically has drift. Even with a very good sensor providing odometry data, there will be drift over time because it is a continuously-updating state based solely on the sensor measurement of change from the previous state. Any error in those measurements gets integrated into the result.

Approaches such as localising to a known map are used to give more robust pose estimates at a point in time, but they tend to have a lower update rate than odometry. So what tools like slam_toolbox do is they integrate the two approaches (or more) together so you get periodic trustworthy updates, and in between you get the higher-frequency estimate from the odometry. The higher-frequency estimate is produced from, for example, taking the most recent lidar/map localisation pose as the baseline and adding on odometry values to it. Because this only needs to work for a short time until the next lidar/map localisation result, the drift is not significant enough to worry about.

There are also additional things that may be going on, depending on the algorithm, such as using the odometry value to constrain lidar-based localisation results, but in general the above is how it works.