Technology / The engineering story

Building physical intelligence for metal recovery.

Four technical questions connect what the system senses, decides, handles and learns. Each capability is assessed against real materials and operating conditions.

Perception / Handling / Identification / FeedbackIn-house core stack
Four questions

How the system works.

Each question is answered by a specific engineering function, then tested against the material task it has to solve.

CAMERAPOSE / GEOMETRYOCCLUSIONSCHEMATIC / NOT TO SCALE
01 / How does it see?

Perception

Machine vision estimates geometry, position and material-relevant visual features, including difficult overlaps and occlusion.

Engineering details +

Camera setup, lighting, coverage, calibration and image labeling are tuned to the selected material task. Specific hardware is configuration-dependent.

BASEGRASP TARGETCONTROLLED PATHOUTPUTSCHEMATIC / NOT TO SCALE
02 / How does it pick?

Robotic handling

Grasp planning coordinates geometry, the robot and end effector, with results informing the next action.

Engineering details +

Actual gripper, cycle path, reachable envelope, safety design and exception handling depend on test results and installation constraints.

VISUAL FEATURESSPECTRAL SIGNALGRADE RULETARGETOTHERCOMBINED DECISIONSCHEMATIC / NOT TO SCALE
03 / How does it identify?

Material intelligence

Visual information can be combined with configured spectral measurements where required and technically available.

Engineering details +

Measurement modality, timing and alloy discrimination must be validated for each specific system configuration and target category.

SORT RESULTLABEL & QANEXT RUNMODEL UPDATEPERMISSIONED DATASCHEMATIC / NOT TO SCALE
04 / How does it improve?

Evidence feedback

Permissioned test and operating data can guide model, process and system improvements over time.

Engineering details +

Define data access, labels, QA protocols and validation gates before using customer material and operating records.

In-house

What we build ourselves.

The core software and integration work stays with our team, so the sorting decision can be tuned to the material rather than treated as a black box.

01 / Software

Perception software

Camera calibration, image processing and feature models for scrap conditions.

02 / Planning

Grasp planning

Target selection, pose estimation and handling-result checks.

03 / Decision

Sorting logic

Combines visual and configured spectral inputs against customer grade rules.

04 / Integration

System control

Robot cell control, safety interlocks and data logging.

Reusable by design

A single application becomes a platform.

Only when the right modules can be reused and the material-specific parameters are explicit.

Reusable

Platform modules

  • Robot cell architecture
  • Perception and calibration stack
  • Grasp-planning software
  • Validation protocol and data schema
Reconfigured

Material-specific parameters

  • Sensing configuration and spectral module
  • End-of-arm tooling
  • Recognition models and labeling
  • Grade rules and acceptance thresholds
Expansion path

A staged expansion path.

Current application, validating applications and future directions are kept separate — status is never implied by a material list.

Current focus

Brass

The first application: mixed yellow-metal scrap into customer-defined brass categories.

Validation in progress
Next evaluation

Copper · Bronze · Aluminum

Configuration requirements and test scope are reviewed before any capability is published.

To evaluate
Longer term

Superalloys · E-waste

Potential evaluation paths that require new sensing or system configuration. Not first-release capabilities.

Future evaluation

Product evidence first. Reusable intelligence next.

Long-term perspective: building physical intelligence for metal recovery.

The company story ↗