Alex CoulombePresents
← the lab

Co-Location Engine

Private — Agile Lens internal, in active development

Fiducial-marker-based co-location for Unreal Engine VR — real and virtual players sharing one room, without a shared coordinate system to begin with.

Multiplayer XR in a physical space has a chicken-and-egg problem: every headset needs to agree on where the room actually is, and nothing hands that to you for free. This system solves it with printed ArUco/AprilTag/QR markers of known position — a camera-relative detection becomes a rigid-body solve that snaps a VR pawn's tracking origin onto real-world map space, then keeps it there as people walk around.

One rigid solve, not a position-then-yaw hack

Position and rotation are solved jointly — a weighted planar rigid-body fit (Kabsch/Umeyama) across every visible marker at once, not a separate 2-marker yaw formula bolted onto an averaged position. More markers in view means a tighter, self-correcting solve instead of diminishing returns from just picking two.

Corrections you can't feel

A locked-in position doesn't get nudged on every noisy frame. Drift correction is velocity-gated — standing still means never moving the player, walking gets an imperceptible pull toward true, and being grossly wrong (inches, sustained across multiple readings) triggers a full snap instead of a slow, visible slide. The goal is co-location that fades into the background.

Three independent detector families

ArUco, AprilTag, and QR all feed the same localizer, each carrying its own reserved ID range so a printed marker map can mix families without collision. A digital-twin overlay renders the exact expected marker pattern at its known position, so misalignment is visible and diagnosable in headset — no guesswork.

Tested without ever putting a headset on

A three-tier headless harness validates the whole pipeline before it ever touches hardware: pure-math unit tests against synthetic marker geometry, a real render-target camera detecting a live 3D scene, and a fully scripted OpenXR device driving the packaged app end-to-end. The last one already closes the loop on co-location accuracy to sub-centimeter — headless, in CI, no one wearing anything.

At a glance

  • Joint N-marker rigid-body solve — yaw and position from every visible marker at once, not sequential 2-marker hacks
  • Velocity-gated drift correction: still = untouched, walking = imperceptible, grossly wrong = clean snap
  • ArUco + AprilTag + QR detection unified behind one localizer, namespaced marker IDs
  • In-headset digital-twin overlay for instant visual alignment checks
  • Three-tier headless test harness — unit, synthetic-camera, and full scripted-OpenXR-device — no physical headset required

Want in early?

Co-Location Engine is in active development. Early-access spots, pilot projects, and collaborations are all on the table.

Re: Co-Location Engine