Abstract:
This paper presents a navigation suite combining 3D lidar, MEMS-IMU sensors, and convolutional neural networks (CNNs) for lunar/asteroid proximity operations. We address GPS-denied environments and real-time trajectory adjustments.
Technical Depth:
Lidar-Inertial Fusion: Iterative closest point (ICP) algorithms merge lidar point clouds with IMU data, achieving <0.1 m position error and <0.5° attitude error in crater-rich terrains.
AI Obstacle Detection: YOLOv8-based CNNs detect boulders (≥0.5 m) with 98% accuracy at 100 m range, enabling 5–10 m/s velocity adjustments.
Fuel Optimization: Reinforcement learning (PPO algorithm) reduces ΔV by 22% during descent/landing by prioritizing smooth trajectories.
Innovation:
A dual-lidar system (16-channel + 128-channel units) enables 0.1–100 m scale mapping with seamless handover, critical for pinpoint landing (<10 m error).
