A VISION-AIDED REAL TIME NAVIGATION SCHEME FOR PINPOINT LANDING ON MARS

Landing a spacecraft within 100 m of the predefined site is required in future mars exploration missions, which largely challenge the navigation performance. In this paper, a vision-aided real time navigation scheme, designed for pinpoint landing on mars, is proposed to meet the required precision. In the scenario, a map with multiple layers of different resolution of the landing area is generated by the mars orbiter before entry, descent and landing (EDL). The state estimation of the lander from the navigation filter is used to determine the scale difference between the descent image and the map to improve the efficiency and performance of image feature extraction. Then, the geometric relationship of the key-points in the image is analyzed based on dilution of precision (DoP), and a simple and efficient method was presented to select a few key-points those affect estimation accuracy most as the inputs of the navigation filter, which largely cut the time-cost of feature description and matching. To obtain the state update, the vision observations are tightly fused with the inertial measurements using an extended Kalman filter (EKF). Finally, the sequential images in the powered descent phase and the map of the landing area are simulated to validate the proposed navigation method. Simulation results show that the mean navigation errors at touchdown are 1.2 m for position, 0.125 m/s for velocity, and 0.082 degree for attitude.