Shigeo Shingo’s original poka yoke concept, developed at Toyota in the 1960s, used physical mechanisms to make errors impossible or immediately detectable. A fixture that only accepts a correctly oriented part, a pin that prevents a component from being installed backwards, a template that signals when a dimension is out of tolerance. These mechanisms work because they remove human judgment from the error-detection step.
The principle has not changed in 60 years. The implementation has. AI poka yoke devices manufacturing in 2026 use computer vision to apply the same logic to process steps that physical fixtures cannot govern: sequence compliance, label verification, presence confirmation, and operator motion patterns.
What is a poka yoke device?
A poka yoke device is any mechanism that prevents an error from occurring or detects it immediately after it occurs. The Japanese term translates literally as “mistake-proofing.” In manufacturing, poka yoke devices are classified by when they act: prevention devices that make the error physically impossible, and detection devices that make the error immediately visible.
Prevention poka yoke examples include asymmetric connectors that only fit in one orientation, height gauges at station entry points that stop non-conforming parts from entering the next process, and software interlocks that prevent a work order from advancing until all steps are confirmed.
Detection poka yoke examples include light curtains that alert when a hand enters a restricted zone, photosensors that detect whether a part is present before a machine cycles, and visual indicators that change state when a process step is completed.
The distinction matters because prevention devices have lower escape rates but higher implementation cost and less flexibility for product changes. Detection devices are more flexible but require fast alert and response mechanisms to prevent escapes.
Traditional poka yoke device categories
Jigs and fixtures. Physical guides that position parts in the correct orientation, making incorrect assembly impossible. Effective for stable, high-volume production but require retooling for product variants.
Photosensors and proximity switches. Detect presence or absence of a component at a defined point. Reliable and low-cost, but only binary: present or not. Cannot assess orientation, completeness, or process sequence.
Force and torque sensors. Detect whether fasteners were driven to specification. Common in automotive assembly. Do not detect whether fasteners were installed in the correct location or sequence.
Error-proofing software. Work order systems that require step confirmation before advancing. Effective for sequence compliance but dependent on operator self-reporting without physical verification.
Vision-based poka yoke: what it adds
Vision-based poka yoke uses a camera and AI inference to check process states that physical devices and sensors cannot govern. A camera covering an assembly station can verify:
- Whether a component was installed (presence check)
- Whether a component was installed in the correct orientation (orientation check)
- Whether a label was applied correctly (legibility and placement check)
- Whether a fastener was installed in the correct hole among multiple options (location check)
- Whether a process step was completed in the correct sequence (sequence check)
- Whether a required quantity of components was installed (count check)
Each of these checks can be performed at production speed, with results available before the assembly moves to the next station.
Nagare implements vision-based poka yoke as part of its digital work instruction and process monitoring platform. The camera observation of each assembly step is combined with the instruction sequence to verify that the physical action matches the required step before the instruction advances.
Which poka yoke approach is right for which situation?
The strongest error-proofing installations combine approaches: a physical fixture governs major orientation, a vision system checks the steps the fixture cannot govern, and a digital work instruction system sequences the operator through both.
