Abstract
This research explores the use of passive imagery on a planetary lander to provided data to a feedback control system. The aim of this innovation is to improve the landing accuracy during the final descent phase. The work formed part of a larger project, the Magnolia mission for a low cost, pinpoint landing system. The gravity turn landing scheme is first investigated. Analytical results are derived to remove the constrains of: 1) the lander having a low initial altitude; and 2) ignoring centrifugal force. A hybrid gravity turn landing scheme based on an angle dependent controller and constant thrust controller is proposed to allows the lander to have an initial altitude up to 70% of a planetary radius and still land safely. The proposed motion estimation algorithm consists of two main steps: 1) Continuous Wavelet Transform (CWT) ridge map extraction, and 2) Random Sampling Consensus (RANSAC) for motion estimation. The CWT ridge maps represent global maps of interesting features. The RANSAC algorithm uses these ridge maps to find the correspondence between frames. The instantaneous motion parameters of the lander’s 3D relative motion with respect to the landing site is then calculated by knowing the initial altitude and attitude when the Optical Navigation system (ONS) is switched on. The dynamic motion model of the lander is therefore estimated using a least square fitting over a block of video. The proposed algorithm is evaluated using a series of videos from lab based experimental setups. Errors under a few percent are achieved. These experimental results show the potential of the proposed algorithm to incorporate an optical navigation system with a feedback controller. However, the system has not been optimized for real time implementation. Therefore, it has been the first item mentioned in the future works.