High-fidelity Synthetic Data Generation for Lunar Exploration via Physically-based Simulation

Ludovica Cavalieri, Mohamed El Awag, Simone Andolfo, Fabio Valerio Buonomo, Riccardo Teodori, Antonio Genova

Keywords: space robotics and automation, data sets for robotic vision, deep learning for visual perception

TL;DR: A Blender-based lunar simulation framework integrating physically-grounded terrain modeling and regolith simulation, photorealistic rendering and automated data annotation for high-fidelity datasets generation.

Abstract: As lunar exploration missions gain renewed interest, driving the need for advanced autonomous systems, the limitations of current methods become evident. The integration of artificial intelligence in the navigation systems of space probes has proven to be a reliable solution to improve the efficiency and robustness of guidance, navigation, and control systems. However, these approaches rely on large, high-quality training datasets, which cannot be easily acquired in space environments. Synthetic datasets have emerged as a viable alternative to mitigate this issue, enabled by the recent advances in photorealistic simulation tools. Nevertheless, models trained solely on synthetic data generally fail to generalize well to real imagery. To mitigate this effect, we developed a unified framework combining physical realism, rendering fidelity, and flexible annotation across multiple mission scenarios using Blender, a 3D modeling suite that supports physically grounded photorealistic rendering. The framework is evaluated on representative space robotics tasks, such as terrain classification and slope computation. The results demonstrate that the simulator can serve as a dataset generator and testbed for advanced perception pipelines, supporting the development of AI-enhanced navigation pipelines for different space exploration scenarios.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Paper Acceptance: No

Submission Number: 20