Jetson Edge AI
Inspection that sets itself up.
A multi-camera AOI system you install yourself. It learns a new SKU in minutes and inspects in any light and any pose, aligned or not.
Machine vision used to take integrators, custom rigs, light boxes and weeks of tuning. Not anymore.
Six things only OnPrem Vision does.
Conventional AOI needs an integrator, a fixture for every part and a light box. OnPrem Vision needs none of them.
Multi-camera, self-installed
One to four cameras, mounted by your own team and running on their own. Fits almost any production line.
Teach a new SKU in minutes
Right in the app, by the person running the line. No vision engineer, integrator or code: the AI picks the settings and trains the model.
Robustness to environmental lightiing
Robustness to ambient lighting variations, eliminating the need for enclosed inspection chambers and enabling seamless integration directly into existing production lines.
Real-time defect and anomaly detection on aligned parts
It learns what a good part looks like and flags anything that deviates, part by part, as the line runs. Teach specific defect classes on top for the ones you already know.
Watch the full demoPose-agnostic detection for parts that arrive any way up
Rotated, flipped, or a shape that can’t stand on the desk or conveyor: it finds the part and inspects it in whatever pose it lands. Taught from a 3D scan of one good part. No fixtures.
Watch the full demosNo-code integration with PLCs and cobots
Connect industry-grade PLCs from a visual workflow. A cobot loads samples from the desk or conveyor and bins them by verdict.
- Modbus TCP
- PROFINET
- EtherCAT
- Cobot load and bin
- Diverter · stack light
See it on real parts.
Real parts, real cameras, the whole app on screen. One aligned-part line, two parts in any pose.
From box to line.
Five steps, all done by your own team. Nobody flies in to install it, and nobody writes a line of code.
-
01
Unbox
Cameras, ring lights, stands and the edge unit ship in one case.
-
02
Point the cameras
One to four, over a conveyor, a desk or a station. No calibration.
-
03
Mark the part
Outline it once per camera. The AI auto-tunes all cameras & lighting parameter for best inspection.
-
04
Collect good samples
The app guides the capture. It learns normal, so no defect samples are needed. The AI picks every setting and trains.
-
05
Run the line
Pass or reject per part, in real time, straight to the PLC.
Plugs into the line you already run.
No code. Drag a PLC link into the workflow, pick your protocol and deploy. The controller gets pass or fail, the part ID, and which camera saw the defect.
-
01 · Load
Cobot picks the part
From the desk or the conveyor belt, and presents it to the cameras.
-
02 · Inspect
OnPrem Vision decides
One to four cameras, one verdict per part, on the edge unit.
-
03 · Signal
PLC gets the result
Verdict, part ID and a defect list per camera, over the fieldbus.
-
04 · Bin
Cobot bins it
Pass and fail bins, or a diverter and stack light on the line.
- Modbus TCP
- Master or slave
- PROFINET
- IO-device, GSDML supplied
- EtherCAT
- Master or slave, ESI supplied
- Compute
- NVIDIA Jetson Thor T4000/T5000
- Capture
- Synchronized views
- Inference
- Local, real time
- Retraining
- Local, on drift
- Network
- Air-gapped
Versus conventional AOI.
- Pharmaceutical
- Medical devices
- Electronics
- Food and beverage
- Automotive
- Plastics
- Metals
- Packaging
Ready when your line is.
Send us a sample part, aligned or not. We’ll show you the result.