SLAM fails in repetitive environments

1. Symptoms

• The map stretches or shortens along a corridor.
• Loop closure joins two different aisles.
• The robot localises to the wrong aisle after a restart.

2. Possible causes

• Scan matching is under-constrained along a featureless direction.
• Loop closure accepts a match between similar places.
• Odometry is the only constraint along the corridor.
• Visual place recognition confuses repeated patterns.

3. Measurements to collect

• Scan-match response or covariance along the corridor.
• Loop-closure candidates and their scores.
• Odometry error over the corridor length.

4. Diagnostic commands

Commands are not executed by the site’s automated tests (they need a ROS 2 Jazzy installation or the robot).

ros2 topic echo /slam_toolbox/graph_visualization --once
ros2 bag record /scan /odom /tf

5. Step-by-step investigation

1. Find the direction without constraint (along the corridor).
2. Add a constraint there: wheel and IMU odometry, a feature at the ends, markers.
3. Tighten loop acceptance (higher match response, smaller search distance) or require a consistency check.
4. Add global cues (a map, markers or GNSS) to disambiguate.
5. Test restart localisation from several start places.

6. Corrective actions

• Improve odometry and fuse the IMU.
• Add reflectors, fiducial markers or a known landmark.
• Raise loop-closure thresholds and add geometric verification.
• Use an initial pose from a known dock.

7. Validation after the fix

• Corridor length in the map matches the measured length within the requirement.
• No false loop closures in repeated runs.

References

• slam_toolbox loop-closure parameters: example configuration, read 2026-11-06.