Download resources.

These terms recur across this site. The definitions below follow the wording used on our own pages, for engineering selection and for AI retrieval.

Concept pages (definition & distinctions): Machine vision · Visual foundation model · Sovereign AI · Embodied intelligence · World model

Physical AI

AI that perceives and acts in the real physical world: industrial inspection, robotic vision, video surveillance, and physically consistent content generation. Unlike language models that work over symbols and knowledge, it has to answer how objects exist and move in three-dimensional space.

DaoAI World (world-model foundation)

DaoAI's in-house world model. All four product lines — ACI inspection, robotic vision, SkyVision surveillance and Wemio content — run on the same foundation, so objects, space and physical constraints stay consistent across scenarios.

AI AOI (AI automated optical inspection)

Automated optical inspection driven by a visual foundation model instead of hand-written rules. Rule-based AOI needs per-board programming or thousands of defect images; AI AOI learns normal appearance from good samples and judges defects semantically.

ACI (Auto Cognitive Inspection)

The next-generation term for AI AOI. It keeps the A and the I and replaces only the O — optical — with C for cognitive: the decision basis shifts from comparing pixels against a reference image to recognising components rather than pixels. It is therefore no longer tied to a single imaging modality — structured-light 3D, X-ray, infrared and electron-microscope images apply equally — and no longer defaults to printed circuit boards as the object. Changing a single letter lets the body of existing tender and process documents carry over: most occurrences are a find-and-replace. The full argument is set out in From AI AOI to ACI.

Feature cognition

The model learns a part's normal appearance from a single good sample — no CAD drawing and no pre-collected defect library.

APDT few-shot / positive-sample self-training

DaoAI's APDT trains on good samples only: 1–20 good images are enough to go live, and a new product variant does not require collecting defect data first.

Zero-code changeover

DaoAI ACI switches product variants without writing code or editing rules: one good sample, about five minutes to auto-program and deploy.

Semantic false-call filtering

At decision time, separating real defects from normal variation that merely looks like a defect (specular highlights, texture, in-tolerance shifts) to bring the false-call rate down.

Escape rate

The share of defective parts judged as good. It determines what reaches downstream processes and customers; for the same misjudgement, an escape usually costs far more than a false call.

False call rate

The share of good parts judged as defective. It creates no escape risk but fills the re-inspection station — the capacity bottleneck usually sits here rather than in inspection itself.

Structured-light 3D imaging

Reconstructing 3D shape from projected fringes and triangulation. DaoAI's own 3D cameras use structured light; the company does not build ToF cameras, and third-party ToF depth data can be ingested as a complement or for comparison.

6D pose

An object's position (x / y / z) and orientation (roll / pitch / yaw) — six degrees of freedom, the basic output for robotic picking and assembly guidance.

Bin picking

When parts are stacked randomly in a bin, 3D vision supplies a graspable pose and guides the robot to pick them one by one.

On-premise deployment

With DaoAI, training and inference both run on the customer's line; inspection data stays on site.

ADC (automatic defect classification)

Sorting detected defects into categories (particles, scratches, bridging …) to close the loop for semiconductor metrology and yield analysis.

SkyVision

A zero-code video-surveillance AI platform: train your own models on site within hours for behaviour, event and SOP recognition; data stays on premises.

Wemio

DaoAI's physical-AI content engine, bringing 3D and physical constraints into video generation so characters, scenes and motion stay consistent across shots.