摘要 : Accurate localization is essential for remote rover in scientific planet exploration missions. However, limited by energy supply, onboard computing capacity often cannot handle all multisensor fusion tasks simultaneously, which de... 展开 Accurate localization is essential for remote rover in scientific planet exploration missions. However, limited by energy supply, onboard computing capacity often cannot handle all multisensor fusion tasks simultaneously, which degrades localization performance rapidly. In this article, we propose a new ground station-assisted visual-inertial (VI) simultaneous localization and mapping (SLAM) approach. Considering the inevitable communication delays of rover-station network, image frame rate would be quite low. Hence, many features tracked in current frame can be easily lost, which causes large tracking errors or failures. To address these challenges, we propose a new spatiotemporal visibility-driven feature selection method. In spatial aspect, optical flow constraints are used for selecting distinguishing features outside the boundary regions. In temporal aspect, we propose a virtual keyframe state generation method, which predicts camera poses in near-future horizons. Therefore, it is conducive for selecting features with the largest visible probabilities by information matrices, which improves tracking robustness and pose estimating efficiency. Furthermore, we design a visibility driven VI-SLAM (VisiVIS) system and build a simulation platform of planetary surface analog environments. Extensive evaluations are performed with different amount of time-delays, and the results verify that our method achieves better localization accuracy and competitive real-time performance. 收起