Onboard Observation Planning for Sequential Image-based Planetary Landing Navigation in Unknown Environments

Abstract

Autonomous navigation based on sequential images is widely accepted as an effective method to guarantee pinpoint landing and precise obstacle avoidance in planetary landings. Observing a large number of landmarks and processing a large number of images places a computational burden on landers with limited computational resources, but existing studies lack observation planning methods for landmarks in unknown environments that can be operated autonomously on landers. In this paper, we analyze the observability of a vision-aided inertial navigation (VAIN) system and propose an observation planning strategy. The minimum number of landmarks and the minimum number of observations are obtained from the observability analysis, which is the boundary condition for observation planning. The landmark configuration planning in the observation planning strategy optimizes only univariate convex functions to efficiently select a small number of landmarks in unknown environments with high navigation accuracy, and the observation interval planning adaptively reduces the number of observations without significantly affecting the navigation accuracy. Simulation results verify the correctness of the observability analysis results, and it is found that the proposed observation planning strategy outperforms traditional observability degrees in terms of both improved navigation accuracy and computational speed when observing a small number of landmarks, and can effectively reduce the number of observations.