Entwicklung eines Partikelfilters zur Lokalisierung mobiler Roboter mithilfe von 3D-Gaussian-Splatting

An RGB‑based particle filter that incorporates a 3D‑Gaussian‑Splatting reconstruction is presented for localizing a TurtleBot 4. The method is evaluated in both simulated and real indoor environments, demonstrating low pose errors for local and global localization and robustness to the kidnapped‑robot scenario, while showing that medium‑trained reconstructions provide the best trade‑off between accuracy, throughput and GPU memory usage. These results indicate that 3D‑Gaussian‑Splatting can serve as an effective renderable map representation for visual robot localization.