Robotic Localization via Active Projected Fiducials for Space Construction

Mengjun Wang1 · Shuai Li1 · Yunlan Zhang2

Journal of Computing in Civil Engineering · 2026

Affiliations

1 Department of Civil & Coastal Engineering, University of Florida, Gainesville, FL, USA

2 Department of Civil, Architectural and Environmental Engineering, The University of Texas at Austin, Austin, TX, USA

01 / THE PROBLEM & RATIONALE

The Problem

Robotic construction in extraterrestrial environments requires reliable localization without GNSS. Conventional odometry accumulates drift, while natural visual features and physically installed landmarks may be difficult to rely on across unstructured terrain.

Our Rationale

Rather than relying solely on existing landmarks, we actively create visual references by projecting coded fiducials onto the terrain. By registering these projected landmarks with a digital elevation model (DEM), we establish known 3D references for camera-based localization without physically placing markers at each location.

Illustration of accumulated localization drift · Conceptual simulation

02 / METHOD

Active Projected Landmark Localization

We develop a terrain-aware localization framework that integrates calibrated projection, digital terrain geometry, and camera-based pose estimation.

  1. Project — Generate Visual Landmarks

    A calibrated projector casts identifiable fiducial patterns onto the terrain, creating temporary visual references.

  2. Register — Establish 3D Correspondences

    Projector rays are intersected with a registered digital elevation model to determine the 3D positions of the projected landmarks.

  3. Localize — Estimate Camera Pose

    A camera detects and identifies the projected fiducials. Their 2D image observations and corresponding 3D positions are used to estimate camera pose through Perspective-n-Point (PnP).

Complete research workflow. Offline: calibrated projector rays intersect the terrain to build a map from point identities to 3D coordinates. Online: the rover camera detects and decodes projected points, then a PnP solver estimates six-degree-of-freedom pose.
Complete offline and online research workflow. Open full-size diagram (new tab).

03 / PERFORMANCE

Localization Performance

Simulation-Based Evaluation

Mean translational error
3.0–5.2 mm
Successful localization viewpoints
48/48
Nominal simulation conditions.

Effect of Terrain Geometry on Localization Accuracy

Using terrain geometry to determine projected landmark positions reduces localization error on curved terrain.

BASELINE — Fitted local-plane
67.1 mm
PROPOSED — DEM-raycast
4.8 mm
Mean camera localization error · Curved terrain · 9 matched simulation viewpoints

Physical Validation

Valid pose estimates
18/18
Differences relative to an AprilTag-based reference
4.4–10.2 mm
Bench-scale laboratory trials.

Research Implications

This research demonstrates the feasibility of creating terrain-registered visual landmarks for robotic localization without physically installing reference markers. The approach provides a foundation for future localization systems supporting robotic construction in GNSS-denied environments, including planetary surfaces.