Before you begin
Starting knowledge
- Know the product and the cost of missed defects and false rejects.
- Have a quality owner who can adjudicate difficult examples.
Equipment & environment
Start offline with representative, permissioned images and a controlled lighting/camera study.
No fixed completion time: use the evidence gates below to decide when to advance. Resource links open the botstest directory entry; check its publisher documentation for the exact version and configuration.
Beginner
Define defects and image evidence
Know what the camera must reveal and how labels are decided.
Learn in this order
- Defect categories, borderline cases and visibility
- Lighting, optics, pose and calibration
- False accept versus false reject consequences
Put it into practice
- Collect normal, defective and ambiguous examples across representative conditions.
- Write labeling and adjudication rules with the quality team.
A defect catalog, imaging brief and annotation guide.
Ready for the next step when…
Resources for this stage
Study image processing and camera-related tooling through its upstream project documentation.
Compare depth sensing against range, scene, calibration and supported software requirements.
Investigate depth and edge-perception hardware in a recorded sensing experiment.
Intermediate
Validate a baseline without leakage
Compare an inspection method using an honest production-like split.
Learn in this order
- Batch/session splits and class imbalance
- Threshold selection and confusion matrices
- Rejected, unobservable and out-of-distribution inputs
Put it into practice
- Train or configure a simple baseline and evaluate held-out batches.
- Inspect errors by defect type and lighting/position condition.
A held-out validation report and threshold decision record.
Ready for the next step when…
Resources for this stage
Study image processing and camera-related tooling through its upstream project documentation.
Inspect point clouds and geometry-processing results before using them in a perception pipeline.
Visualize spatial observations and compare them with model and trajectory outputs.
Professional
Integrate decisions and monitor drift
Make inspection actionable and traceable on the line.
Learn in this order
- Part traceability and reject handling
- Latency, image quality and drift alarms
- Revalidation after camera, supplier or product changes
Put it into practice
- Run a shadow-mode pilot before automated disposition where appropriate.
- Rehearse uncertain-result escalation and validate after a controlled process change.
An inspection acceptance pack and monitoring/change-control plan.
Ready for the next step when…
Resources for this stage
Inspect recorded robot signals and build views that expose timing and system failures.
Evaluate edge compute against the workload, power, thermal and software requirements you measured.
Investigate depth and edge-perception hardware in a recorded sensing experiment.
Common pitfalls
- Randomly splitting images from the same capture session.
- Optimizing average accuracy while hiding costly missed defects.
Where to go next
Extend to new products only after reviewing defect definitions, imaging and held-out evidence.
Explore other industry paths →