Autonomous Navigation and Safe Landing of an Aerial Vehicle in a GPS-Denied Martian Environment

Objectives: To develop and evaluate a real-time, low-power multi-sensor fusion framework that performs localization and safe landing-zone detection together for an aerial vehicle operating over Martian terrain, where neither GPS nor prior maps are available. Method: A stereo camera (Intel RealSense D435i), a rotating 16-beam LiDAR and an inertial measurement unit (IMU) are fused on a Raspberry Pi 5 and Pixhawk flight controller. LiDAR-assisted visual-inertial SLAM (ORB-SLAM with an Extended Kalman Filter) provides localization, while a classical computer-vision pipeline in OpenCV (CLAHE normalization, Sobel flatness analysis, mask fusion and geometric filtering) selects landing zones. The system was tested in 50 AirSim descent trials from 12 m altitude over Martian-analogue terrain with craters, rocks and dust-covered plains. Findings: Localization error remained below 4 cm, with drift under 0.6 m over 100 m; visual-inertial fusion cut drift by 45% compared with vision-only tracking. Map coverage reached 95%. The LSDS classified no-fly zones with 93% accuracy, needed 18.5 ms per frame (about 5 times faster than the CNN-based HALO approach) and guided the vehicle to within 0.3 m of the target. Landing succeeded in 98% of trials (49/50; 95% CI 89.5–99.6%) at under 10 W, against more than 20 W for GPU-based VINS-Mono at comparable accuracy. Novelty: Earlier GPS-denied SLAM and planetary landing systems address either localization or hazard assessment or depend on GPU/LiDAR-heavy hardware. This work runs both functions concurrently on a single sub-10 W embedded platform without training data.