When parts lie flat on a conveyor or carrier, their height may already be known. With height and tilt controlled, a calibrated 2D camera can supply the position and rotation needed for picking. Learn which assumptions must hold, how height variation affects the result and how to test the approach.
Required condition
The reference plane, part height and permitted tilt are known.
Decisive test
Check independent positions at the image edges and the permitted height extremes.
Reconsider when
Height or tilt breaks the placement tolerance: constrain presentation or measure additional geometry.
The known plane must correspond to the geometry used for localization. It might be a flat part's top surface, or a visible reference surface with a known offset to the grip. A conveyor at a fixed height is insufficient if parts vary in thickness or can tilt. Define the permitted product variants, height range and orientation before selecting the sensing method.
The robot usually receives X and Y coordinates and rotation about the plane normal. The remaining orientation and approach height come from the process recipe. A part that can tip over needs additional sensing or a presentation method that prevents that condition.
An image point defines a viewing direction. Knowing the working plane allows that direction to be intersected with the plane. This requires the camera model, lens-distortion treatment and the plane's relationship to the robot. With a suitable imaging model, a planar transformation, also called a homography, can describe the mapping. A single arbitrary pixels-to-millimetres factor is generally insufficient across the image.
The detected part position is not yet the commanded tool position. The offset from the detected feature to the grip, tool length and tool centre point belong in the same coordinate chain. Check transformation direction and units as well as the numerical result.
If the real surface lies outside the calibrated plane, the same viewing direction intersects the assumed plane at a different location. A Z error can therefore produce an X or Y error. Its size depends on camera angle, position within the image and the optical model. An oblique camera deserves particular attention.
Adding pixels does not correct a wrong plane assumption. Options include constraining presentation, validating a separate plane for each product recipe, or measuring height and tilt. Select the approach against the permitted placement error and the consequences of an incorrect pick.
| What do you vary? | What do you check? |
|---|---|
| Position and rotation | Use independent points at the centre and edges. Measure position error and final placement. |
| Height and tilt | Test permitted extremes and an out-of-range part. Check when the system rejects the observation. |
| Motion during the cycle | For a moving conveyor, measure delay to the pick and test compensation for part travel. |
Record detection, positional error, grip outcome and final placement separately. Set acceptance limits for the relevant stages. A stationary demonstration does not validate a moving conveyor: measure the delay between acquisition and picking and validate motion compensation where needed.
Compare camera, optics, lighting and software together with presentation hardware, calibration effort and changeover time. Mechanical guidance may simplify sensing while adding space and maintenance requirements. Include rejects, recovery and operator interventions. A payback calculation needs measured frequencies and times rather than an assumed picking success rate.
Share a part image, height range, permitted tilt, placement tolerance and cycle time so we can define a focused trial.
Discuss your application