Align robot sensor time without hiding assumptions

Evidence and interpretation

rosbag2 distinguishes publication and reception ordering during playback. Maha’s proposed alignment record preserves both when available and records any clock conversion used for analysis. Missing offsets should not be guessed from visually similar events. When clocks reset or messages arrive late, divide the record into intervals whose assumptions remain valid. A normalized timeline is a derivative, so retain the original timestamps and the conversion revision beside it.

Proposed evidence workflow

1. Identify clock domain, timestamp units, reset events and whether each timestamp denotes capture, publication or receipt.
2. Document measured offsets and bounds on alignment uncertainty; refuse an unsupported conversion.
3. Recompute event comparisons after alignment changes and retain the previous result.

Worked illustration

A synthetic camera frame arrives after a release acknowledgement but was captured before release. Ordering solely by arrival would suggest the object remained held afterward. The proposed report keeps the ambiguity until capture and acknowledgement clocks can be related.

Limits

No clock synchronization service or real-time guarantee is implemented. Reordering a recording does not repair missing observations.

Sources and review

Continue reading

• Robot evidence needs explicit frames and units
• Replaying robot logs is not reproducing an experiment