LunarSite: An end-to-end ML pipeline for lunar south pole landing site selection

Short summary

LunarSite is an open-source ML pipeline that ranks lunar south pole landing sites by combining terrain segmentation, crater detection, and XGBoost ranking across 315k grid cells. Trained entirely on synthetic data, it achieves 0.8456 mIoU and transfers to real NASA imagery with strong sim-to-real generalization and calibrated uncertainty estimates. Results validate against NASA's Artemis III candidate regions with 81-85% agreement on deep-shadow terrain.

• •End-to-end ML pipeline for lunar landing site selection with live demo and open-source code
• •Three-stage architecture: U-Net terrain segmentation, crater detection fine-tuned on LOLA DEM, XGBoost site ranking
• •Trained on synthetic Unreal Engine data; validates against NASA Artemis III regions with strong sim-to-real transfer

Generated with AI, which can make mistakes.