3D vision + robotics · Kraków, Poland

Pick & place driven by a CAD file

We build a system that recognizes parts on a production line straight from STL/CAD files and tells the robot what to grasp and how. No tooling built around each part, no neural-network training — changeover means uploading a file.

contact: kontakt@dudarobotics.pl

loaded: Oasis_040125-001.STL
STL model of a sheet-metal part — isometric view
The real part — bent galvanized sheet metal on the test rig
part: recognized6D pose: OKgrasp point: g1validation: ADD-S
no per-part training real-time on a CPU any 3D camera any ROS2 robot accuracy report per part
How it works

From an STL file to a grasp — five steps

The same CAD model you already have from design or from your customer is enough for the station to start working with a new part.

01 / input

STL / CAD file

You upload the part model — no photo sessions, no data collection.

02 / profile

Part profile

The system generates the geometry, grasp point and an accuracy validation report.

03 / scan

3D camera

A point cloud of the scene — conveyor, table or bin.

04 / pose

6D pose engine

CAD geometry matched to the point cloud (PPF + ICP) with a confidence score.

05 / grasp

Robot picks

Pose and grasp point go to the robot — pick and place.

Who it’s for

Two ways to work with us

Manufacturers

Automated part picking

  • Picking parts off the conveyor after bending, stamping, casting or machining
  • Sorting mixed parts against their CAD references
  • Machine loading and short-run packing
  • Viable even at hundreds of pieces and frequent changeovers
Request a free feasibility check →
Integrators & distributors

A 3D vision engine for your cells

  • A container with an API / ROS2: point cloud in → 6D pose + grasp point out
  • Part profiles generated from your client’s STL files in minutes
  • Flexible per-station or per-project licensing — the margin stays with you
  • Engineering support straight from the makers
Book a demo on your client’s parts →
A bent, textureless sheet-metal part — a hard case for classic machine vision
Technology

Geometry over appearance

Works on difficult parts

Smooth, metallic, bent, textureless surfaces — exactly where classic 2D vision fails. The system matches the 3D geometry from the CAD file to the point cloud, so color, glare and scratches stop mattering.

Training-free

No neural-network training for each part and no GPU on the station. The 6D pose engine runs in real time on an ordinary industrial PC.

Measured, not promised

Every part goes through accuracy validation (an ADD-S report) before it reaches the station. You know what to expect before you invest.

Open architecture

Any 3D camera and any robot controlled over ROS2. A part library in the cloud or on-premise, with full data isolation for every client.

Pilot program

Start with four low-risk steps

Step 1

Feasibility check

You send STL files and photos of your parts — we assess the fit and point to the best first operation.

free of charge
Step 2

Proof of concept

Your parts on our test rig, with an accuracy report and a recording of the system at work.

Step 3

Pilot on your line

A cobot-and-3D-camera station on a chosen operation, with clear success criteria.

Step 4

Scale-up

New parts and operations are just new STL files — no station rebuild.

We play with our cards on the table: the technology is implemented and tested on our test rig, and we deliver the first deployments as partner pilots — on preferential terms, with clear go/no-go criteria after every stage.

Contact

Let’s test your parts

Send 1–2 STL files and a few photos of parts from your production. We’ll come back with a feasibility assessment and a proposed next step. We usually reply within 48 hours, and we’re happy to sign an NDA.

kontakt@dudarobotics.pl

DudaRobotics · Paweł Duda · Kraków, Poland