Common Requirements and Troubleshooting
About 1421 wordsAbout 5 min
7. Common Requirements and Troubleshooting
7.1 Improve Repeat Localization Stability
- First, fix the image capture position, camera parameters, and incoming target reference, then capture the same aluminum ingot repeatedly.
- Check whether changes in the point cloud are caused by reflections, vibration, or occlusion.
- Check whether the template consists mainly of repeated planes and lacks a stable orientation constraint.
- Compare features before and after sampling to ensure that the localization structure does not disappear because of downsampling.
- If an orientation prior is needed, it must be based on the actual constrained orientation of the incoming target.
When using aluminum ingots of different models or with different structural relationships, create and verify separate templates. Do not treat an offset measured once as a universal compensation for all models.
7.2 Multiple Marking Locations on One Target
Multiple marking locations on the same aluminum ingot do not mean identifying multiple aluminum ingots in one operation, nor do they automatically mean “capture once, pick multiple.”
If multiple fixed locations on the same target must be processed after one localization, first define the relationship between the localization reference and each process location. Then verify how the vision output, equipment-side offsets, and execution sequence are implemented. Reuse that localization result only when the target remains stable throughout the process and all locations and responses to exceptions have been verified.
Do not enable caching by default or apply the depalletizing pick sequence in this case. For parameter behavior, see the Vision Calculation Configuration User Guide.
7.3 Optimize Cycle Time
Record the total time first, then identify the specific bottleneck:
| Bottleneck | Check first | Do not sacrifice |
|---|---|---|
| Acquisition time | Exposure, acquisition mode, and imaging conditions | Point cloud completeness of the localization structure |
| Preprocessing time | ROI size, invalid point clouds, and number of functions | Valid features and no-target detection |
| Coarse matching time | Number of template points, sampling parameters, and orientation constraints | Ability to match the correct position and orientation |
| Fine matching time | Number of target point cloud points, search mode, and radius | Final localization accuracy and stability |
| Saving time | Whether all debugging data is continuously saved | Information needed to trace exceptions |
After basic acceptance checks pass, evaluate ultra-fast mode or custom saving as needed. After each optimization, retest the actual localization error and the no-target state; do not compare log times alone. For details, see Vision Parameter Configuration.
7.4 No Localization Result
- Confirm that the current task is “Plane type Target positioned assembly (only Point Cloud matches).”
- Check whether the camera captured the target and whether the correct ROI and template are selected for the current task.
- Check whether the ROI or preprocessing removed valid features.
- Review empty ROI detection and related logs to rule out a false empty result caused by too few points after downsampling.
- Distinguish among coarse matching failure, fine matching failure, and localization points being filtered out.
- Correct the corresponding stage. Do not adjust 2D segmentation scaling or confidence parameters that do not exist in this task.
For related methods, see the Point Cloud Matching Pose Estimation Configuration Guide and Empty ROI Detection Configuration Guide.
7.5 Matched to the Wrong Plane or a Repeated Structure
- ROI contains multiple similar locations: limit it to the current target and the allowed deviation, and exclude adjacent targets.
- Template contains only a plane: add a stable boundary or a distinctive combination of structures.
- Features are degraded by reflections or occlusion: improve acquisition first and, if necessary, choose a new localization reference.
- Incoming target orientation varies too widely: verify the conditions under which the scene is applicable; configure an angular constraint only when a real prior exists.
- Coarse matching lands at the wrong location: correct the features, ROI, and coarse matching first. Do not accept an incorrect result by relaxing fine matching or the score threshold.
For how to recreate the template, see Point Cloud Template and Marking Localization Point Configuration.
7.6 Localization Point Offset or Incorrect Orientation
7.6.1 Incorrect in the Vision View
- Check whether the template is aligned with the correct structure.
- If the template is correct but the localization point is wrong, check how the target localization point is defined relative to the template.
- Verify whether the template was recentered, rotated, or replaced without updating the localization point.
- Check whether runtime pick point adjustment adds another offset.
- If the template shifts slightly, compare edge noise, missing features, and point clouds before and after sampling.
7.6.2 Correct in Vision but Incorrect on the Equipment
- Verify the eye-hand calibration and current installation state.
- Verify the TCP, units, rotation representation, and coordinate system conventions.
- For an eye-in-hand setup, verify that the image corresponds to the robot pose at the time of acquisition.
- Check whether the equipment applies the working distance or a fixed offset twice.
- Check whether a historical result or a cached result from before the target moved is being used.
First distinguish among vision registration, coordinate transformation, and process execution errors. Do not use a single compensation to correct all sources at once. See Checking and Analyzing Calibration Results.
7.7 Pose Available but Localization Point Filtered Out
- Confirm that the target localization point has been saved and associated with the current task.
- Identify the first filtering function that excludes the result and review its reason.
- Check whether the localization point lies within the specified ROI type region; a location near the matched feature is not necessarily an allowed output location.
- Check the ROI axis direction, positive and negative directions, and angular constraints on the tool pose axis.
- Check the matching score and tool collision model configuration, and validate them separately with normal and abnormal samples.
For handling methods, see the Pick Point Filtering Function Configuration Guide. Do not restore automatic marking by disabling necessary filters.
7.8 Output When No Target Is Present
- Check whether supports, conveyor lines, or other similar structures remain in the ROI.
- Compare point counts with and without a target; do not equate “enough points” with the presence of the target.
- Check whether the template can produce a false match against the background.
- Verify the combined effect of empty-state detection, match quality, and output-region filtering.
- Check whether the equipment treats the previous successful result as valid for the current operation and continues execution.
The execution program must explicitly handle no-target and failure states. It must not continue marking after merely displaying an error in the interface.
7.9 Issue Materials and Privacy Check
For reproduction, gather the software version, generic sample ID, template and parameter versions, historical data and logs from the same operation, and describe the expected position, actual position, and conditions under which the issue occurs.
Before sharing externally, check the text, screenshots, filenames, and attachment directories. Remove customer names, personal information, project codes, internal addresses, and account credentials. Surface markings on aluminum ingots, handwritten numbers, QR codes, and actual marking content may also contain business information; replace them with approved, nonsensitive examples. Do not provide unconfirmed site images, target models, or logs as public downloads.
Return to Case Overview · Previous Chapter: Operation and Acceptance