Degradation is the default
A long straight pipe wall supplies weak geometry: every scan looks like the same slice of circle. Wet walls, mist and dust further remove useful LiDAR and vision points. Repeating texture sends visual matching to the wrong correspondence. The localisation system may still output a pose. That pose may no longer be trustworthy.
Public methods such as LIO-SAM and FAST-LIO2 are strong learning baselines because they engineer laser-inertial fusion. Reproduction is not about copying the same trajectory plot. It is about testing when to alarm, how to take over after the alarm, and how logs let a person replay.
The product should promise an alarm, not mythical accuracy
PipeGuardian writes 0.2–3 m wall-hold as its own control-mode target, and writes drift plus degradation alarms in long-straight / wet-wall / dust as a freeze gate. The specs page does not write centimetre figures in the present tense. Autonomous-exploration scores from large mine voids do not transfer to a long narrow pipe.
Segment-level evidence does not need perfect global coordinates. It needs travel from the last elbow, sensor time alignment, and a person who can find the matching frame in replay. Over-promising absolute accuracy drags G3 into an argument that never closes.
How to read this back to PipeGuardian
This article is industry education, not a pilot news item. If a vendor, university, funding scheme or standard appears, treat it as a contrast question: what must our own rig answer, and what must public copy forbid. Copying an external figure into a PipeGuardian spec sheet conflicts with the evidence page.
The English site carries the full method set: system architecture, safety bounds, target specifications, the 18-month roadmap and HK–SZ operations. Field work still starts with a permit and a recovery path. Reading an article is not a confined-space permit.