Mag-MM: Magnetic-Enhanced Multisession Mapping in Repetitive Environments

Multisession mapping is becoming increasingly crucial for a wide range of applications in robotics and autonomous systems, including surveying, map updating, and multiagent collaboration. However, current solutions either rely on the matching of elementary geometric or semantic features, which typically leads to failure in GNSS-challenged degenerated environments with repetitive features (e.g., office/hotel/hospital long corridors, warehouses, gardens, tunnels, and underground facilities); or rely on a good prior estimate/input of the initial transformation matrices of multiple single-session maps, which may not always be obtainable. Characterized by its omnipresence and distinct variability across different locations, the ambient magnetic field is well-suited for estimating the initial transformation matrix regardless of the geometric features. Therefore, we present Mag-MM, a novel probabilistic framework for magnetic-enhanced multisession mapping that addresses the aforementioned challenges. The proposed framework aims to estimate the transformation matrices of maps from multiple single sessions, and to build consistent and accurate global maps in repetitive environments. Our framework introduces three key innovations: A comprehensive probabilistic framework that addresses the problem of multisession mapping in repetitive environments; a particle swarm optimization (PSO)-based magnetic data association strategy to autonomously determine the initial estimate of the transformation matrices; a magnetic fingerprint correspondences-based algorithm to delimit the overlapping regions for data redundancy reduction and efficiency improvement. Results of simulations and real practical missions show the significantly enhanced mapping consistency and accuracy over the baselines.