Aura Scan Pipeline

Each link below reflects an evidenced family, lineage or direct connection. This project has 7 relevant public connections.

Aura interface, geometry and capture architecture

Phone scans in, measurable CAD-ready twins out, processed entirely on your own machine. You walk around a picnic table, a wall or a skate bowl with an iPhone, export the scan from a LiDAR app (LiDAR is the phone’s laser depth sensor: it measures distance with light the way a bat does with sound), drop the file onto a Windows PC, and one command turns it into web-ready 3D models, a point cloud and a measured, layered CAD drawing. No cloud services, no accounts, no uploads. The data never leaves your own gear.

Public page: Aura Scan Pipeline

If you remember nothing else

Four commands. Everything else on this page is detail.

aura-scan init my-scan --mode asset      # 1. make a folder
                                          # 2. drop the phone export into scans\my-scan\raw\
aura-scan run scans\my-scan               # 3. process it
aura-scan report scans\my-scan            # 4. see what came out

Results land in scans\my-scan\exports\: a light 3D model for the web, a full-detail one, a point cloud, a dimensioned CAD drawing, and a record card saying what was scanned and how far the measurements can be trusted.

The one optional step worth doing: tape-measure two distances on site, then run aura-scan pick scans\my-scan to click those same two points on screen and write the real numbers into manifest.yaml. That is what turns a scan that looks right into a drawing an engineer can trust. Full instructions are in docs/capture-guide.md.

Where this is up to

Plain status, no spin:

Who it serves

Quickstart

Needs Python 3.12 and uv (winget install astral-sh.uv if you don’t have it). From the repo root:

uv venv
uv pip install -e ".[dev]"
.venv\Scripts\activate

Worked example, a picnic table:

# 1. Make a scan project (creates scans\picnic-table-one-mile\)
aura-scan init picnic-table-one-mile --mode asset

# 2. Scan the table with a phone app and drop the export
#    (OBJ, PLY, GLB, LAS or a raw depth package) into
#    scans\picnic-table-one-mile\raw\

# 3. Optional but recommended: click the corners you tape-measured on site
aura-scan pick scans\picnic-table-one-mile

Then open scans\picnic-table-one-mile\manifest.yaml and add the tape measurements against the picked points:

control:
  picked_points:            # written by aura-scan pick
    - [0.412, -0.788, 0.731]
    - [2.109, -0.815, 0.742]
  distance_constraints:
    - {points: [0, 1], distance_m: 1.685, note: "table top, long edge"}

# 4. Run the full pipeline
aura-scan run scans\picnic-table-one-mile

# 5. Re-read any processed scan later
aura-scan report scans\picnic-table-one-mile

Everything lands in scans\picnic-table-one-mile\exports\: meshes, point clouds, a dimensioned DXF drawing, a metadata.json sidecar, a thumbnail and a web viewer snippet.

Three capture modes

Mode For What it adds
general Any geometry Balanced defaults, automatic plan and cross sections
asset Public assets (picnic tables, toilets, bbqs, halls, skate features) Oriented dimensions, footprint outline, condition field, inventory-ready simplification
projection_surface Flat or near-flat surfaces for projection art Plane fit, flatness check, usable-area outline, a plain suitable / not suitable verdict

The LOD ladder

One capture, four tiers of detail, produced in a single run. LOD means level of detail: the same object at different weights for different jobs.

Tier What you get Good for
LOD0 Oriented bounding box and 2D footprint A map pin and an inventory row
LOD1 Low-poly web mesh (GLB, under 15k triangles) Mind palace objects, browser model viewers, the places catalogue
LOD2 Full-detail cleaned mesh (OBJ, PLY) The archival geometry
LOD3 Engineering pack: point cloud (PLY, LAS), dimensioned DXF plan and sections, flatness and dimension reports AutoCAD-compatible work and tenders

Two capture lanes

More to read

🤝🔷 A Luke × Claude build. Public Source Licence.

Luke Hayes × Claude, Minjerribah · 2026.