Vision-Guided Arm: A Passing Simulator Is Not Physical Completion
Separates the verified software loop and firmware build from acceptance work that requires real hardware.
Project: STM32 Vision-Guided Robotic Arm Sorter
Symptom: Detection, queueing, protocol, controller simulation and firmware builds pass, but no physical arm was available to measure real grasp accuracy.
Hypothesis: Software-loop evidence verifies algorithms and interfaces but cannot replace servo-offset, link-dimension, camera, FSR and electrical acceptance.
Investigation: Ran 14 automated tests, a GUI smoke test and the firmware cross-build; checked ELF/HEX/BIN/MAP plus Flash/RAM reports; then separated every parameter and H01-H10 check that requires hardware.
Root cause: The software model uses configured parameters, while geometric error, supply capacity, sensor thresholds and assembly variation exist only on the real mechanism.
Fix: Report status as software verified, firmware builds, hardware pending; provide staged acceptance from separate power and one joint through four-point calibration, FSR calibration and automatic sorting.
Lesson: When hardware is unavailable, retain reproducible software evidence and state the physical acceptance conditions instead of hiding the boundary.