Multipath-Based SLAM Exploiting Extended Object Estimation and Classification

Abstract

By leveraging geometric and probabilistic information contained in multipath components (MPCs), multipath-based simultaneous localization and mapping (SLAM) enables the localization of both mobile agents and a varying number of map features (MFs). Traditional solutions assume that each MPC is associated with a single MF, while focusing only on MFs’ positions. However, advancements in communication technologies provide higher-resolution multipath parameters (MPPs), resulting in large MFs generating multiple MPCs. This challenges the existing association assumptions and provides opportunities to estimate the extents and shapes of MFs. In this paper, we first integrate the many-for-one association relationship and random matrix-based extent modeling into the existing Bayesian SLAM framework. We then categorize MFs by shape, developing multiple shape and measurement models for each category. By exploring these models, we derive the joint posterior distribution and represent it using a factor graph, which serves as the foundation for our proposed message passing algorithm. Numerical results demonstrate that the proposed algorithm achieves superior localization and mapping performance, successfully classifying different types of MFs while estimating their orientations and sizes.