A Confidence-Aware Fusion Layer for Hazard Assessment in Autonomous Lunar Landing

Autonomous lunar landing requires a spacecraft to identify safe landing areas using camera and sensor data. This is difficult because lighting, shadows, and sensor noise affect how the terrain is interpreted, and AI-based hazard detectors can produce uncertain or incorrect predictions. When hazard detection is reduced to a simple Safe or Hazard label, confidence scores and disagreement between detectors are not used in the final decision. This paper proposes a Smart Confidence Fusion Layer placed between the AI hazard detectors and the final landing decision. The layer takes four inputs: hazard prediction, confidence score, detector agreement, and terrain information. Separate detectors cover rocks, craters, and slopes or rough terrain. When the detectors agree, confidence in the result increases; when they disagree, the site is classified as Uncertain instead of being accepted directly. The final output has three levels, Safe, Uncertain, and Hazardous, which may support actions such as continuing the landing, re-checking the area, slowing down, or avoiding the location. The work is a conceptual framework that has not been implemented on a spacecraft, and no experimental results are claimed.