Occasional odometry glitch

[gestom]

When going through the data from Y3 deployment, we found an issue that seems to originate from the odometric subsystem. According to the logs from Y3 deployment at AAF, there were 6 events when the robot got lost due to a probable bug in the odometric module of the robot. All of these happened in the same area:

At AAF, there is a double wing door with a steel frame, which is on the floor. You can see it at 2:16-2:18 in this video: https://www.youtube.com/watch?v=tmYJ1DMwF2M. When the robot passes through that door, its odometry occasionally (in 6 cases out of ~270) jumps by ~10m or more. Normally, I would attribute this to a wheel skid or wheel stuck, but this glitch affects only x,y position, but not the orientation published over the /odom topic.

Moreover, while the robot forward velocity reported by /odom normally corresponds to the change in the robot x,y position, it does not correspond to it in these cases. The reported speed is sometimes unrealistically high, but still far away from ~100m/s that would correspond to the sudden jump in robot position.

The relevant logs and rosbags of these events are available here:
https://lcas.lincoln.ac.uk/owncloud/index.php/s/qrZ1nAsGpPIt1I6

Any idea what could be the cause of that ?