Localization inputs
What the localization engine needs to find the objects depends on how it recognizes them:
CAD-based localization (CAD Localization, CAD Picking) needs a 3D model of the object. You upload an STL file made in CAD software, or you scan the real object and turn the scan into a mesh in the Studio.
AI-based localization (AI Localization, AI Picking) needs a neural network trained to recognize the objects. You select an existing network, or a network retrained for your objects. No 3D model is needed.
CAD-based localization |
AI-based localization |
|
|---|---|---|
Input |
A 3D model of the object |
A trained neural network |
How you get it |
Upload an STL file (CAD file), or scan the real object and edit the mesh (Mesh tools) |
Select an existing network, or a network retrained for your objects |
Where in the Studio |
The Picked object step of the solution |
The Neural Network setting of the localization profile. See General settings. |
Engine |
For the solution types and their setup steps, see Solution types.
CAD-based localization: a 3D model
The CAD localization engine compares the 3D model with the point cloud from the 3D sensor to find each object and its pose. You can give the Studio the model in two ways:
CAD file – upload an STL model made in CAD software.
Mesh tools – scan the real object with the 3D sensor and clean up the scan in the Studio.
You load and edit the model in the Picked object step of the solution.
Which model to use
CAD file |
Mesh tools |
|
|---|---|---|
Source |
A full 360° STL model made in CAD software |
A scan of the real object, taken from the sensor’s point of view |
Ideal for |
Smaller rigid objects picked from a bin in random orientations |
Bigger objects that are semi-oriented and always picked from about the same direction, for example doors on carriers |
Why |
The object can lie in any orientation, so the engine needs the model from every side |
The sensor always sees the same side of the object, so a one-sided model from a scan is enough |
You can also use the mesh tools on an uploaded CAD file, for example to fix its normals or move its origin.
AI-based localization: a neural network
The AI localization engine does not use a 3D model. A neural network finds the objects in the scan instead, so AI solutions have no Picked object step.
You select the network in the Neural Network setting of the localization profile. See General settings. The networks you can use depend on the software modules of your installation:
Existing networks for common objects: boxes (AnyPick Module for Boxes (Depalletization)), totes (AnyPick Module for Totes (Depalletization)) and slipsheets (AnyPick Module for Slipsheets).
Retrained networks for your own objects (AnyPick Module (Custom)).
See Software modules.
Contents
- Requirements for an uploaded STL model and how to upload it.
- Scanning the object and editing the mesh: cropping, simplification, filtering, normals and origin.