Skip to content
Manual anchoring

Manual anchoring

ManualAnchoring is the simplest setup: one source scan with one anchor, one client scan trying to find it.

Scene setup

  1. Create a GameObject with a ManualAnchoring component.
  2. On Client A (source: places the anchor):
    • Assign the scanned root Transform (the parent of all MeshFilter objects) to the Source Root field.
    • Assign an anchor Transform to the Anchor field, or leave it unset to use the source root’s own position as the anchor.
    • Call TryCreateAnchor() (or use the sample’s UI) to extract an AnchorPayload and save it to a file/send to the other device.
  3. On Client B (target: finds the anchor):
    • Assign the scanned root Transform to the Client field.
    • Load the AnchorPayload that Client A sent.
    • Call Calculate(payload) to run registration.
    • On success, Client B’s scene graph is rotated/translated so the anchor lands at Client A’s anchor position in world space.

Inspector parameters

Most projects use defaults.

SectionParameterDefaultMeaning
Feature ExtractionAnchor Radius5 mHow far from the anchor to collect source features
Max Target Features3000Cap on client-side feature budget (scales to device mesh complexity)
Curvature Min0.05Minimum curvature to include a vertex as a feature
MatchingCurv Diff Max0.1Max curvature difference for a feature to match
Fine-Tier ICPRefine Local ICPtrueSecond-stage anchor-local refinement after global registration
Fine Radius0 (auto)Radius around anchor for local refinement; 0 = auto-calculate
Room TierBuild Room TiertrueExtract whole-scan features for better coarse-stage hints and room fingerprinting
DescriptorUse Gravity DescriptortrueInclude gravity direction in descriptor (improves robustness on most devices)

The algorithm is gravity-aware (uses the device’s IMU, its built-in motion sensor) and scale-relative: thresholds adapt to mesh density and coarseness automatically. Start with all defaults.

If your anchor is on a flat surface (desk, wall), lower Anchor Radius to 1-2 m. That makes features cluster at distinctive corners and edges rather than sampling blank space.

If registration fails on your data, see Diagnostics for how to read the log: it reports feature counts and which verification gate failed.

Next