Knowledge · Robot vision

2D robot picking: a fixed-plane worked example

Flat parts lie separately on a stationary carrier. They can shift and rotate, while their height remains fixed. This illustrative example works through the steps from a camera image to placement in a fixture. It provides a trial structure, without claiming customer results or measured production performance.

Explicit assumptions

The carrier is stationary, height is known and parts do not overlap.

Test the whole chain

Evaluate detection, grip offset, grip confirmation and final placement separately.

Example status

This is an illustrative trial structure without invented customer results or savings.

Worked example: the complete picking chain
Explanatory schematic, not a production capture or measured result.

State the assumptions

Assume each part rests on the same reference plane, tilt is constrained and parts do not overlap. The camera is fixed and the carrier remains stationary from imaging to picking. A visible feature has a known relationship to the desired grip point. A real application must test these assumptions against product variation.

The destination is a fixed fixture. Required placement tolerance, permitted contact pressure and maximum cycle time are customer inputs. This example does not assign fictional performance figures to them.

Build the coordinate chain

Calibrate image coordinates to the working plane and relate that plane to the robot base. Detection supplies a reference point and rotation. Rotate the known grip offset with the part before computing the tool target. Z follows from the validated plane and the product and tool geometry.

Use a simple test to confirm that an image displacement produces the expected robot direction. Then test several orientations. A sign error or exchanged axis can remain hidden at one convenient position.

Design acquisition, gripping and placement separately

Choose illumination that keeps the relevant feature visible across permitted surfaces and ambient conditions. Reject unsuitable images. Then check that the proposed grip has adequate contact and clearance. Approach from a validated position and confirm acquisition using tested feedback.

Placement requires its own geometric check. Final placement error can exceed localization error if the part shifts in the tool. Verify the physical outcome, not only the calculated camera coordinates.

Test boundaries before optimizing

Start with independent positions at the centre and edges. Vary rotation and use the extremes of permitted part height. Include a tilted part and an unrecognizable part to check exception handling.

Trial stageResult to record
Detection and localizationRetain the image, detected position and measured position error.
Grip and confirmationRecord the selected grip and whether the agreed confirmation signal is correct.
Placement and cycle timeMeasure fixture placement and elapsed time, including failure handling.

Investigate failures before increasing motion speed. Repeat the relevant acceptance tests after a speed change.

Compare complete alternatives

If height variation cannot be controlled, a fixture or different infeed may be needed. Depth sensing is an alternative where the task genuinely needs additional spatial information. Reassess tooling, camera position and cycle time with that change.

Compare investment, engineering, operating labor, maintenance and measured rejects with the current process. This example provides a trial structure, not predicted savings. Evidence from the real trial determines which approach proceeds.

Work through your own picking task

Bring sample parts, presentation variation, placement requirements and the intended operator role so we can turn them into explicit assumptions and acceptance criteria.

Discuss your application