A Measurement Scheduling Method for Multi-vehicle Cooperative Localization Considering State Correlation

Abstract

Accurate and robust localization is a fundamental requirement for safe autonomous driving. Currently, most vehicles rely on single-vehicle localization (SVL) using the global navigation satellite system (GNSS) for obtaining the position data. However, in urban canyons or areas with signal interruptions, the GNSS signals may be unreliable, thus leading to accumulated errors in SVL. In contrast, cooperative localization provides the advantages of superior accuracy, good fault tolerance, and high flexibility by leveraging multi-source data from self-sensors and neighboring vehicles. Nevertheless, establishing communication links with all surrounding vehicles can introduce a severe communication burden and thus impact real-time data processing performance. To address these challenges, this paper proposes a cooperative localization method with loosely-coupled communication that considers the state correlation. In this study, the communication process is divided into two topologies to ensure filter consistency, and an approximate variable substitutions technique enables suboptimal estimation of a cross-covariance matrix. Furthermore, a dynamic period measurement scheduling (DPMS) method is proposed. The DPMS optimally selects vehicles for cooperative localization based on the covariance matrix trace values, considering three factors of location uncertainty and employing a weight function for period adjustment. Finally, the proposed method is evaluated through multiple simulations and field tests, assessing the algorithm convergence, localization accuracy, execution time, and communication burden. The results demonstrate that compared to the sequential greedy algorithm, the proposed method can significantly reduce communication and computation burden while achieving higher localization accuracy than the random measurement scheduling method and the measurement scheduling method based on a deep neural network.