Operation and Acceptance
About 1341 wordsAbout 4 min
6. Operation and Acceptance
Acceptance is divided into three stages: reproducing results with historical data, live capture, and low-speed robot integration. Proceed to the next stage only after the results of the previous stage are correct. A software result alone does not establish that the robot can pick the target.
6.1 Establish a Baseline with Historical Data
- In the visual computation configuration, switch Input Data to Historical Data.
- During debugging, select Save All (Debug Mode) and save the settings.
- Select a complete historical dataset representative of normal incoming cylinders, and confirm the corresponding target, ROI, template, and vision parameters.
- Click Run, then locate the results and logs by the time of this run.
- Save the configuration versions and results from before and after the changes, and use the same input to compare the effects of parameter tuning.
If no suitable historical sample is available, first capture and save a normal scene with the camera while prohibiting automatic robot motion. Then use that data to establish the baseline.
For instructions, see the Historical Data User Guide and Visual Computation Configuration User Guide.
6.2 Inspect Vision Results by Stage
6.2.1 Inspect 2D Instances
- Does each cylinder requiring handling correspond to a separate instance?
- Have adjacent cylinders been merged, or has a single cylinder been identified more than once?
- Are the edge positions, directional features, and surfaces actually visible captured completely?
- Have the tray, fixture, partition, or background been treated as targets?
- Do the count and numbering after filtering and sorting match the incoming material for this run?
A correct box around one cylinder does not mean that recognition is correct across the entire scene. Inspect every position requiring handling.
6.2.2 Inspect the Point Cloud and Pose
- Select an instance and check whether its Point Cloud comes from that cylinder and whether it contains points from adjacent targets or the background.
- Check whether the template’s outer contour and features that distinguish its orientation fit the instance.
- Confirm the pose from multiple angles. In particular, check the front and back, ends, and symmetric orientations; do not judge from the top view alone.
- Compare the Coarse Matching and Fine matching results. If the deviation increases after Fine matching, inspect the point pairs, template region, and background interference.
- Capture the same cylinder repeatedly and check whether its pose jumps between captures.
For highly symmetric targets or targets with insufficient usable features, a high matching score may still leave the orientation ambiguous. Judge the result against both the orientation and positioning error required by the actual process.
6.2.3 Inspect Pick Points and Filtering
- Are the Pick Points at the intended gripping positions, and does the approach direction match the actual TCP?
- When the cylinder’s position and angle change, do the Pick Points transform correctly with the target?
- Are valid Pick Points incorrectly filtered out? Are results with abnormal positions, angles, or collisions allowed through?
- Does the final output order meet the process requirements? When a strict order is required, does a failure at one position trigger an alarm as required?
- Do the number of poses calculated and the number of instances sent meet the test objective? Avoid mistaking an output count limit for missed recognition.
6.3 Use Logs to Locate the Processing Stage
When more detailed processing information is needed, switch the log level to DEBUG and compare only records from this run.
| Stage | Key information | Check first if an exception occurs |
|---|---|---|
| Image and instance generation | Instance count, segmentation, and filtering results | Model, ROI, imaging, scaling, and Confidence |
| Instance sorting | Grouping, sorting, and retained results | Coordinate system, axis direction, step size, and extraction range |
| Pose estimation | Whether Coarse Matching and Fine matching succeeded | Corresponding keypoints, template alignment, and instance Point Cloud |
| Pick Point processing | Number generated, reasons for filtering, and output results | Pick Point of Target Object, angle, ROI, matching score, and collision conditions |
For complete instructions, see Log Description. Before exporting logs or screenshots for external communication, check project names, file paths, device addresses, and other sensitive fields.
6.4 Retest with Live Capture
- Switch Input Data back to Camera Data and save the configuration.
- Arrange the cylinders as they arrive in actual operation, covering the permitted variations in position, orientation, and height.
- Trigger a run and inspect the 2D results, Point Cloud, pose, and Pick Points as described above.
- Save separate samples for normal, edge, missing-target, empty, and permitted occlusion conditions.
- If the current capture results differ noticeably from the historical data, first check whether the Camera, lighting, target condition, or installation conditions have changed.
6.5 Low-Speed Robot Integration Acceptance
Prerequisites for Integration
The Robot may use vision output only after the coordinate system, units, pose representation, and calibration results have been confirmed. The scene in historical data may differ from the current physical setup and should not be used directly to execute a pick.
- Confirm that Camera Data is the current input and that the installation state of the target and fixture matches the selected configuration.
- Review the calibration results, number of robot axes, pose representation, TCP, and output coordinate system.
- Confirm that the test area is clear of people and obstacles, that stopping conditions are in place, and that debugging follows on-site safety procedures.
- Use low speed, single-step, or single-cycle operation to first verify arrival at a safe approach position above the target.
- After confirming that the approach direction and positioning error meet the process requirements, verify gripping, lifting, withdrawal, and placement.
- Extend the test to edge positions, different orientations, and continuous incoming material. Verify that incorrect or outdated vision results are not used under abnormal conditions.
For calibration checks, see Checking and Analyzing Calibration Results. For communication and coordinate conventions, see the Robot Configuration Guide.
6.6 Acceptance Records and Pass Criteria
Specific accuracy, Cycle time, and test quantities should be determined by on-site process requirements. This case does not provide a single set of pass values that can be applied directly.
| Test item | Results to check |
|---|---|
| Normal incoming rows and columns | Separate recognition, correct poses, and stable picking |
| Edge positions and permitted position deviations | The ROI does not crop key features, and positioning and picking still meet requirements |
| Permitted orientation and height variations | No flips or obvious pose jumps; order meets requirements |
| Missing or no incoming targets | The background is not treated as a target, and outdated pick results are not returned |
| Occluded or unpickable state | Filter or alarm according to the configuration; do not attempt an abnormal pick |
| Continuous operation | Accuracy, failure handling, and Cycle time remain stable under the specified test conditions |
Record sample numbers, configuration versions, results at each stage, positioning deviations, pick results, elapsed time, and exception handling. Use generic numbers in external presentations; do not include customer or production-line identifiers.
Return to Case Overview · Previous Chapter: Vision Parameter Configuration · Next Chapter: Common Requirements and Troubleshooting