SLAM-Based Navigation Scheme for Pinpoint Landing on Small Celestial Body

Abstract

Addressing the need for robust pinpoint landing capabilities, this paper proposes a monocular navigation scheme based on the Rao-Blackwellized particle filter simultaneous localization and mapping (SLAM) approach. The proposed online navigation scheme provides attitude and position (or pose) estimation during the approach, descent, and landing phase for small celestial body missions. This approach relies on one navigation camera and potentially sparse readings from one or more range sensors (e.g. LIDAR (Light Detection And Ranging)). The concept of the proposed navigation scheme is to maintain several hypotheses of the most likely spacecraft pose and landmarks position and to feed the most likely one to the spacecraft controller at any given time. The proposed system uses a double staged Monte-Carlo simulation that represents the population of all possible spacecraft motions between two camera images taken at successive time steps, and that samples this population over all possible scaling factors, converting each relative motion to world-scaled coordinates in the process. The purpose of this Monte-Carlo based visual pose estimation approach is to offer an alternative solution to the drift error and inaccuracy problems of SLAM kinematic models, odometry motion models, and other conventional dead reckoning techniques.