I’m working on an offline monitor for AMR deployment traces. In simulation, I’ve seen multi-sensor state signals change before a modeled failure boundary.
Has anyone seen something similar in real fleets? What signals do you look at before docking misses or localization resets?
I’m speaking a bit hypothetically here since I don’t know the algorithms you’re using or the details of your system. I might be able to help more if we discuss a specific case.
For localization, you could consider developing a localization health monitoring system. Of course, the signals you’d want to monitor would depend on the localization algorithm you’re using.
For example, if you’re using graph-based localization, you could periodically score how well the current sensor observations match the existing map or graph as the robot moves. By tracking how these scores change over time, you might be able to detect changes in environmental features or potential degradation in localization quality before an actual failure occurs.
For odometry and IMU, you’d probably need different approaches. There are already several algorithms for detecting IMU anomalies. Gyroscope measurements can also be affected by temperature, so monitoring temperature changes and sensor error codes could be useful as well.
For docking, I’d suggest starting by splitting the process into two stages: dock detection and dock approach. I’m assuming you’re using some kind of pattern detection to identify the dock. Similar to the localization approach, you could score how well the detected pattern matches the reference. This would let you analyze whether successful docking attempts show similar scores or patterns over time, and how those scores change before failed attempts. At the very least, this could help you understand whether the problem starts during detection or the approach phase.