What you calibrate
The rigid (or periodically re-checked) relation between camera frame and robot flange/base.
Knowledge · Robotics
Hand-eye calibration: linking camera and robot frames.
Vision reports positions in camera coordinates; the robot moves in base coordinates. Without stable calibration the grasp misses, even with a perfect detector.
Concept
Eye-in-hand vs eye-to-hand
Eye-in-hand: camera on the robot/wrist. Eye-to-hand: fixed camera viewing the cell. Both need a calibrated transform; maintenance and drift differ.
Why it fails
Loose mounts, tool change without recal, temperature drift, wrong TCP.
Diagrams
Lab photos and annotations help, this page stays conceptual until Vision Lab visuals are available.
Practice
What to include in the project
Calibration is not a one-time checkbox. Plan repeatability, tooling and who may re-calibrate.
-
01
Lock mounts mechanically
Play in camera or tool mounts kills calibration faster than software bugs.
-
02
Document the procedure
Targets, poses, acceptance criteria and when to re-calibrate.
-
03
Tie to the grasp
Calibration error + model/detection error stack, budget both.
-
04
Test real grasps
Validation is a missed or successful pick under line conditions, not only an RMS number.
Calibration in your cell
Tell us whether the camera is fixed or on the wrist, we outline a feasible calibration path (visuals when available).