Fiducial Marker Splatting for High‑Fidelity Robotics Simulations

High‑fidelity 3D simulation is critical for training mobile robots, but traditional approaches relying on mesh‑based representations struggle in complex environments with dense foliage, occlusions and repetitive structures. Recent neural rendering methods such as Gaussian Splatting achieve photorealistic results, yet they lack a flexible mechanism for incorporating fiducial markers that are essential for robotic localization and control. In this work we propose a hybrid framework that combines the photorealism of Gaussian Splatting with explicit marker representations. Our core contribution is a novel algorithm that efficiently generates Gaussian‑splat based fiducial markers (e.g., AprilTags) within cluttered scenes. Experiments demonstrate that the proposed approach outperforms classical image‑fitting techniques in both efficiency and pose‑estimation accuracy. We further validate the framework in a greenhouse simulation environment, where dense foliage and occlusions highlight the benefits of seamlessly integrating fiducial markers into high‑fidelity radiance field renderings.

Recommended citation: D. Tabaa and G. Di Caro, "Fiducial Marker Splatting for High‑Fidelity Robotics Simulations," arXiv preprint arXiv:2508.17012, 2025.
Download Paper