On this page
straighten
Calibration Issues
Residual will not settle, edges stay wrong, or the result disappears after a power cycle.
Residual grows instead of settling
The fit is fighting bad input. In order of likelihood:
| Cause | Check |
|---|---|
| A reference point was placed in the wrong position | Re-place the targets exactly where prompted |
| A target moved between capture and fit | Nothing should touch the scene during the run |
| Geometry selection wrong for the surface | A curve modelled as flat fits the centre and fails the edges |
| Point cloud itself is unstable | Go back to First point cloud |
Residual good, edges wrong
The centre fits and the corners do not. This is a geometry or coverage problem, not a numerical one.
| Check | Fix |
|---|---|
| Active area larger than the sensor’s coverage | Add a sensor, or reduce the area |
| Sensor sees the surface at a grazing angle | Reposition so the scan hits closer to perpendicular |
| Reference points clustered in the middle | Spread them to the edges of the active area |
Calibration lost after power cycle
The profile was not written to the device. Save it explicitly, then power cycle once and re-check before considering the job done.
Repeatability looks fine but drifts over days
Something is moving. Brackets creep, projectors get knocked, tripods settle. Re-run calibration and compare the residuals — a slow drift points at mechanics, not software.
Next
Working again? Re-run the acceptance tests: Interaction test.