Advanced optics
Advanced Optics System
In-house optical design that images high-reflectance, transparent, curved and anodised surfaces cleanly. What the algorithm cannot see is, in the end, what the optics failed to produce.
Exhibiting ROBOTWORLD 2026 · November 4–7, 2026 · KINTEX Hall 3–5, Goyang, Korea Digital catalogue → NEWDSROBOT industrial humanoid launched →
We build the algorithm, the optics and the machine —
DSTEK KOREA
We build the algorithm, the optics and the machine —
so we can own the whole line
Handdle AI, trained on 100 billion industrial defect images, paired with optics and equipment we design ourselves, right through to automation and DSROBOT, the industrial humanoid that knows inspection. One vendor, end to end.
- Deep-learning inspection
- Custom optics
- Equipment build
- Robotics & automation SI
- Industrial humanoid
- MES/PLC integration
Leading manufacturers — including 100+ Fortune 500 companies — run DSTEK
Customer stories →DSTEK AT A GLANCE
An AI that gets stronger as the data accumulates
From Suzhou in 2019 to 15,000+ production lines across 30 countries. These are numbers from volume production, not from a lab.
- 15,000+ Production lines
- 30+ Countries served
- 100+ Fortune 500 customers
- 100B+ Training images
- 2B+ Images analysed daily
- Up to 99.99% Detection accuracy Application-specific
DSTEK group worldwide
DSROBOT
NEW · Industrial humanoid
A robot that sees, judges and handles
DSROBOT puts two arms on DSTEK’s AI vision — an industrial embodied-AI robot, launched in four bodies: fixed, wheeled, folding-waist and lift.
- MATE
- UNI
- FLEX
- CARRY
Why DSTEK
Including the difficulty other vision vendors decline
Four differentiators that follow from owning the algorithm, the optics and the machine.
Robotics + optics
Robotic Optical Inspection
Irregular three-dimensional parts are covered completely by pairing a robot arm with the optical system — geometry a fixed camera cannot reach.
Data-driven AI engine
Data-Driven AI Engine
Handdle AI, hardened on field data from 15,000+ production lines and 100 billion+ training images. The longer it runs on your line, the sharper it gets.
End-to-end turn-key
End-to-End Turn-key
Equipment development, installation, line integration and operation — software, hardware and automation from a single supplier.
Turn-key
One vendor, accountable to the end
We started as an AI software company, then built our own optical hardware, then the inspection machines, and now the robots, transfer systems, line automation and the data layer above them. Owning every layer means accountability never splits.
- AI software platform Handdle AI — a no-code deep-learning vision engine with MLOps, unifying optical, image-processing and deep-learning algorithms in one platform.
- Optical imaging system We design the cameras, lenses and lighting ourselves — for high-reflectance, transparent, curved and anodised surfaces that other vision vendors turn down.
- Equipment & automation From AI-AOI machine build to transfer systems, robot integration, MES/PLC connectivity and packaging automation — we design and install the whole line.
- Data intelligence Turning inspection results into production data — edge-cloud analytics, PB-scale traceability and an industrial vision foundation model.
- AI smart camera All-in-one vision systems with edge compute built in. Inference completes inside the camera, with no PC or frame grabber.
Solutions
Representative industries
Proven on 15,000+ production lines across 30 countries. Those below are representative — the actual footprint is considerably wider.
Semiconductor & Substrates
Wafer, package and bonding through to PCBA, FPC and FC-BGA.
10 inspection tasks
Automotive & Parts
Stamping, welding, painting and assembly, plus core component inspection.
12 inspection tasks
Power Equipment & Cable
Multi-face inspection of motors, windings, terminals, cable and steel.
10 inspection tasks
Secondary Battery
Electrode, assembly and formation — from foil intake to finished cell.
12 inspection tasksVersus traditional AOI
Where rule-based AOI gives out
Shift the lighting slightly, or introduce a new defect type, and a rule-based algorithm has to be rewritten. A deep-learning model is simply trained.
Traditional AOI
- Rule-based — sensitive to position, angle and lighting
- A new defect type means an algorithm update
- Struggles to classify complex textures such as BGA and QFN
- Over-detection sends the work back to a human anyway
Handdle AI
- Deep-learning — robust to lighting and workpiece variation
- Learns a new defect type from 5 samples — or a single one for large parts
- Classifies complex textures and materials
- Pixel-level segmentation quantifies defect area and position
Solder defects (insufficient, excess, bridging, cold joints, voids) and appearance defects (contamination, flux residue, pinholes, solder balls) are judged by the same engine.
- Up to 99.99% Detection accuracy Application-specific
- From 1 μm Detection precision Depends on the optical configuration
- Up to 200 units/s Inspection speed Depends on workpiece and hardware
- 5 images typical · from 1 for large parts Minimum training samples Depends on the part
- 30+ Transfer-learning compatible models
- 1 year Warranty maintenance included Per contract
These are maximums achieved on specific applications. Real performance depends on the part, the optics, the model and the hardware — a PoC establishes the figures for your conditions.
Featured stories
DSTEK running on the floor
Electronics & 3C — inspection machine
Earphone-housing inspection machine
- 16Inspection stations
- 900 UPHThroughput
- 0.02 mmSmallest defect identified
Batteries — EVE Energy
Cylindrical lithium cells, every process across seven plants
Automotive — vehicle OEM
Vehicle paint, inspected end to end by robot and AI
Customer references
The manufacturers that lead their industries run DSTEK
Leaders in automotive, batteries, electronics, displays and photovoltaics run DSTEK inspection in volume production — on 15,000+ lines in over 30 countries, including more than 100 Fortune 500 companies.
