The robot repeatedly misses the part by a small amount. Is the error in the camera measurement, the coordinate transform or the tool? Independent test positions help separate those causes. Use this guide to measure the offset, investigate its pattern and check it against the placement requirement.
Measure independently
Use verification points that were not used to solve the calibration.
Examine the error pattern
Compare systematic error and repeat spread by position, height and tool orientation.
Acceptance
Check physical placement against the task requirement using suitable measurement uncertainty.
First establish what is wrong. Is the estimated object position incorrect, does the tool reach a different point than expected, or does the part move during gripping? These observations require different investigations. Camera intrinsics describe image formation. Camera-to-robot calibration relates coordinate frames. Tool calibration defines the working point relative to the robot flange.
Robot repeatability concerns the spread of repeated motions. It is different from absolute agreement between commanded coordinates and physical position. A vision system can also return a repeatable but biased estimate when it uses an incorrect height or grip offset.
Use identifiable reference points or a measurement fixture with known uncertainty. Keep verification positions separate from the data used to solve the calibration. Check a printed target's physical dimensions, flatness and mounting. The digital drawing does not establish the dimensions of the printed board.
Approach the target through the normal process chain. Measure the resulting offset with an appropriate method rather than using a fragile tool or product as a hard stop. Record measurement uncertainty; a verification method coarser than the required tolerance cannot demonstrate compliance with that tolerance.
Distribute test points throughout the usable work area. Include different heights and tool orientations where the application requires them. Repeat positions to distinguish scatter from systematic offset. Retain the image, robot position, recipe version and relevant conditions for each measurement.
Report error by direction and orientation where relevant. An average alone can hide a failure near an edge. Examine the largest observed error and repeat spread as well. Passing the test set does not establish performance outside the tested region.
| Observed pattern | Focused check |
|---|---|
| Nearly constant offset | Check origins and fixed offsets in the coordinate chain. |
| Error rotates with the tool | Check the tool centre point and the grip offset. |
| Error grows towards the image edge | Investigate lens distortion and the assumed working plane. |
| Error changes with height | Check the spatial model and transforms to robot coordinates. |
These patterns guide investigation; they are not definitive diagnoses. Change one factor at a time and repeat the same independent checks. Recalibration can temporarily compensate for an error without resolving its mechanical cause.
Repeat relevant checks after moving the camera or robot, changing lens settings, a collision, or changes to tooling and mounting. Decide whether startup or changeover needs a shorter routine check. Retain a traceable combination of calibration, tool data and product recipe.
Include checking time, recovery and downtime in maintenance planning. Acceptance limits should follow the placement task and gripper requirements rather than a general accuracy claim for one component.
Bring the work area, required placement tolerance and examples of the observed error. Specify whether the camera is fixed or moves with the robot.
Discuss your application