Elevation-Aware Deep Learning Framework for Safe Lunar Rover Path Planning Using Chandrayaan-2 Datasets

Navigating the lunar surface, especially in the south polar region, is challenged by irregular topography, steep slopes, and crater abundance. By employing Chandrayaan-2 datasets to locate craters, this work suggests a deep learning-based method to solve the issue of finding safe and ideal navigation routes for lunar rovers. TMC2 ortho-images and TMC2 DTM datasets are used to achieve 2D and 3D terrain knowledge, while high-resolution OHRC images are used to train YOLO models for precise crater recognition. Crater detection serves as a critical preprocessing step in the proposed system, enabling the planner to avoid hazardous zones and compute a scientifically optimal and safe path for rover navigation. A specially designed PyQt-QGIS application imports the identified craters as georeferenced shapefiles, which are then superimposed on the lunar surface. The application includes an$\mathbf{A}^{\boldsymbol{*}}$pathfinding algorithm that uses a cost function that combines elevation variation, crater avoidance, and Euclidean distance to determine the best navigation routes. In order to guarantee realistic rover motion planning, the cost function favors uphill penalties. YOLOv8m was found to be the best model for crater detection after a comparative analysis of several YOLO variations. With a final precision of 79.19%, recall of 26.21 %, and$\text{mAP} {@} 0.5$of 32.34 %, the YOLOv8m model demonstrated its suitability for incorporation into selfgoverning lunar navigation systems. The proposed system can be extended to other planetary bodies like Mars, making it reusable across interplanetary missions.