Abstract: To meet the intelligent requirements of the lunar rover in complex environments, enhancing the decision-making capabilities of agents through training is of critical importance. Currently, for lunar missions there remains a lack of high-fidelity simulation environments capable of accurately reproducing complex terrain and lighting conditions. There is also an urgent need to construct diverse training data that covers extreme scenarios encountered during long-distance exploration. This achieves a joint simulation of "lunar terrain scenarios, equipment models, and environmental perception", establishing a high-fidelity training support system for lunar surface mobile intelligent agents. Firstly, utilizing digital moon technology, a centimeter-level precision lunar terrain model was constructed, providing a physically accurate foundation for agent training. And an integrated simulation scenario coupling lunar environment and rover dynamics models was established to simulate the entire workflow of lunar surface operations. Lastly, a real-time bidirectional data exchange links were established between the rover's perception, navigation, planning, and control algorithms and the simulation platform, enabling continuous learning and evolution of the intelligent agent within a closed-loop simulation. This system provides a critical technological platform for training autonomous capabilities and validating algorithms for mobile lunar rover agents, offering essential data support. It holds significant engineering value for the implementation of China's manned lunar exploration missions.

Key words: manned lunar exploration, lunar rover, joint simulation, intelligent agents, lunar environment simulation