Spacecraft Fiducial Marker for Autonomous Rendezvous, Proximity Operations, and Docking

Abstract: Robotic operations in space are challenging due to the harsh environment and the high cost of failure. Fiducial markers provide visual references that aid autonomous rendezvous, proximity operations, and docking for space robots. However, existing fiducial markers are mostly single-scale and largely designed for terrestrial robotics. Such markers leave the camera's field of view at close range, precisely during the proximity and docking phases where reliable tracking is most critical. This paper presents AstraTag, a fiducial marker designed for autonomous on-orbit robotic operations. The marker template is based on a square Spidron pattern whose recursive, self-similar structure enables detection across multiple spatial scales. Marker identification uses a 48-bit signature derived from triangular sub-regions of the template and encoded with a Generalised Reed-Solomon (GRS) code. The detection pipeline performs contour-based quadrilateral localisation, perspective normalisation, and signature matching against a pre-computed dictionary. To handle markers affixed to curved spacecraft surfaces, it incorporates a Thin-Plate Spline (TPS) re-warp fallback that exploits the marker's internal rectangular borders as additional geometric correspondences. We benchmark AstraTag against three-layer Fractal ArUco and AprilTag on spacecraft mockups with flat and curved surfaces. On curved surfaces, AstraTag achieves a higher detection rate than both baselines, offering a robust recursive-marker option for space robotics.

• The paper introduces AstraTag, a recursive Spidron-based marker with error-corrected 48-bit encoding, multi-scale detection, and a thin-plate spline fallback for curved spacecraft surfaces.
• AstraTag achieved 97–100% detection from 10° to 70° out-of-plane rotation on a cylindrical mockup, while Fractal ArUco fell to 0% beyond 40° and AprilTag reached 56% at 70°.
• The results support spacecraft applications in servicing, assembly, and debris removal, while highlighting the need for equal-area benchmarks, algebraic decoding, and orbital hardware-in-the-loop validation.

AstraTag is a fiducial marker designed specifically for autonomous rendezvous, proximity operations, and docking (RPOD) in space robotics, addressing a gap left by terrestrial marker systems: single-scale markers leave the camera's field of view at close range, precisely during the phases where tracking is most critical. The paper introduces a recursive, Spidron-based marker template with error-corrected encoding, a detection pipeline that handles curved spacecraft surfaces via thin-plate spline (TPS) re-warping, and a comparative benchmark against Fractal ArUco and AprilTag on spacecraft mockups (2606.27566).

Motivation and design context

The work is situated in the context of in-space assembly and manufacturing (ISAM), on-orbit servicing (OOS), and active debris removal (ADR), all of which require accurate relative state estimation between chaser and target. While markerless vision-based pose estimation is possible, it demands information-rich imagery that is difficult to obtain under space-like illumination—extreme shadow/illumination contrast, strong reflections, and bright sources against dark backgrounds. Fiducial markers offer a mature complement, and the design aligns with the emerging AIAA S-155 draft standard developed with CONFERS, which defines baseline requirements for high-contrast spacecraft fiducials including unique identification, crisp edges, and 6-DoF pose estimation. Prior space deployments include the Space Vision Marker System for Shuttle–ISS rendezvous, NASA Astrobee, Altius DogTag grappling fixtures, and commercial docking plates from Astroscale and Orbit Fab.

Existing multi-scale options have notable drawbacks. Fractal Markers nest ArUco codes recursively, but each layer carries a different code, inflating dictionary size, and square-grid encoding limits information density per layer. AprilTag supports tiled flexible layouts but remains fundamentally single-scale per tag. AstraTag instead replicates the same encoded pattern at reduced scale within the marker center.

Marker geometry and encoding

The template derives from the square Spidron, a self-similar structure of alternating triangles adapted to a square perimeter so that the outer quadrilateral contour can serve perspective-n-point pose estimation. Recursive inscription of a polygon through side midpoints scales each nested square by $\sigma_4 = 1/\sqrt{2}$ and rotates it by $\pi/4$, so each layer has half the area of its parent ($s_k = s_0\,2^{-k/2}$). This recursive property enables multi-range detection: when the outer layer becomes unresolvable at close range, the inner sub-marker takes over.

Encoding proceeds by subdividing each of eight triangular regions into six non-overlapping triangles using seven control points (three vertices, three edge midpoints, an approximate centroid), yielding 48 binary cells. The signature is structured as a Generalised Reed–Solomon (GRS) codeword, providing error correction against local image degradation; dictionary entries store all four cyclic-rotation signatures together with world corner coordinates, trading dictionary size for detection speed and requiring cooperative access to the target's dictionary.

Detection pipeline

Detection runs in two stages. First, CLAHE contrast normalization and Gaussian adaptive thresholding produce a binarized image robust to illumination gradients. Hierarchical contour retrieval captures the nested structure; candidate quadrilaterals are validated via Probabilistic Hough line segments grouped into parallel pairs—a step that recovers plausible quadrilaterals even with incomplete contour closure—and refined with Douglas–Peucker approximation and sub-pixel corner fitting. Second, a perspective homography normalizes the patch, and each signature bit is determined by area-based sampling over triangular cells, integrating hundreds of pixels per cell for noise robustness. Identification uses Hamming-distance matching against the dictionary. When standard detection fails on non-planar surfaces, a TPS re-warp fallback exploits the marker's internal white rectangular borders as additional geometric correspondences to correct combined perspective and cylindrical distortion. Pose estimation uses IPPE, a non-iterative analytical planar PnP solver that returns two hypotheses resolved by reprojection error.

Experimental results

Benchmarks used two mockups under simulated space lighting: a cylindrical BAS-01 space station module (curved surface) and an Aditya-L1-style flat-faced bus wrapped in golden insulation. All markers were specified at $10\times10$ cm tagsize, though the authors note an important caveat: AprilTag's tagCustom48h12 layout extends data cells outside the nominal boundary, producing prints $1.67\times$ wider per dimension and $2.76\times$ larger in area than the other two markers.

On the curved surface, the headline results are:

Condition	AstraTag	Fractal ArUco (3L)	AprilTag (48h12)
Out-of-plane rotation, 10°–70°	97–100%	Collapses to 2% at 40°, 0% beyond	100% to 60°, 56% at 70°
Mean detection over distance	87.2%	21.9%	100%

All three fail at 80° rotation, where the marker approaches edge-on and contour-based quadrilateral recovery breaks down. Fractal ArUco's collapse stems from its inner layers requiring substantially more pixels to decode under compounded cylindrical distortion. On the flat surface, curvature effects vanish: AstraTag and AprilTag achieve 100% detection in all conditions, and Fractal ArUco recovers to 99.3% mean over distance, confirming that surface curvature—not algorithmic weakness—is the primary differentiator.

Two additional results merit emphasis. A Metalphoto anodized-aluminum print (sealed inside the oxide layer for thermal and UV survivability) achieved 100% detection and identification at Hamming threshold $\tau_H \le 5$ across the full distance range at $1280\times720$ resolution, failing only at $\tau_H=3$ and 30 cm due to specular glare. With a low-resolution $320\times240$ camera, AstraTag sustained near-perfect recall up to 75 cm regardless of Hamming threshold, degrading gracefully at 100 cm.

Limitations and open questions

The authors are candid about several constraints. The AprilTag area discrepancy means the distance-range comparison confers roughly 2.5× more pixel coverage on AprilTag; the authors state explicitly that an equal-area experiment is required to assess algorithmic efficiency fairly. AstraTag's mean distance performance on the curved surface (87.2%) trails AprilTag's 100%, dropping to ~49% at 150 cm where recursive inner layers become too small to resolve. Identification currently relies on exhaustive dictionary matching rather than algebraic GRS decoding, which the authors plan to replace; storing all four orientations trades dictionary size against detection speed and restricts operation to cooperative scenarios where the servicer holds the target's dictionary. IPPE provides only local approximations for non-planar targets. Finally, evaluation is confined to scaled ground mockups with a single point light source; no orbital or hardware-in-the-loop validation is reported.

Conclusion

AstraTag contributes a well-motivated recursive fiducial design whose self-similar Spidron structure, GRS-protected 48-bit encoding, and TPS-based curved-surface fallback extend reliable marker detection to conditions where existing systems fail—most notably high out-of-plane incidence angles on cylindrical spacecraft surfaces, where it maintains 97–100% detection through 70° versus complete failure of Fractal ArUco beyond 40°. The benchmark also clarifies that flat-surface performance is not a meaningful discriminator among modern markers. Open questions include equal-area benchmarking, algebraic decoding to replace dictionary lookup, and validation under representative orbital illumination dynamics.