G-Loc: Tightly-coupled Graph Localization with Prior Topo-metric Information

TL;DR

G-Loc addresses robust map-based localization by leveraging prior topo-metric information through a tightly-coupled pose-graph framework that fuses LiDAR-inertial odometry, cloud-to-map registration, and GNSS. The method maintains a sliding active graph that integrates LO constraints, map-derived matches to a subset of the reference graph, and GNSS priors, while dynamically loading submaps to handle large environments and GNSS-denied regions. Key contributions include exploiting the topology of the prior model for efficient search and uncertainty handling, a complete modular pipeline with real-time performance on ROS2, and validation across urban datasets and a real autonomous bus deployment. The results demonstrate cm-level localization accuracy, scalability to large maps, and robustness in mapless areas, making G-Loc suitable for real-world autonomous navigation where prior maps exist but are partially available or intermittently accessible.

Abstract

Localization in already mapped environments is a critical component in many robotics and automotive applications, where previously acquired information can be exploited along with sensor fusion to provide robust and accurate localization estimates. In this work, we offer a new perspective on map-based localization by reusing prior topological and metric information. Thus, we reformulate this long-studied problem to go beyond the mere use of metric maps. Our framework seamlessly integrates LiDAR, inertial and GNSS measurements, and cloud-to-map registrations in a sliding window graph fashion, which allows to accommodate the uncertainty of each observation. The modularity of our framework allows it to work with different sensor configurations (e.g., LiDAR resolutions, GNSS denial) and environmental conditions (e.g., mapless regions, large environments). We have conducted several validation experiments, including the deployment in a real-world automotive application, demonstrating the accuracy, efficiency, and versatility of our system in online localization.

Figures (7)

• Figure 1: Visualization of G-Loc in a sequence of the Newer College Extension dataset zhang21. The prior topological (orange pose-graph) and metric (colored pointcloud) model is exploited for online robust localization. An active graph (blue) is optimized, containing certain vertices from the prior graph and the most recent robot states.
• Figure 2: Block diagram of G-Loc. The dashed lines represent the initial guess fed to the registration methods.
• Figure 3: Scheme of how the different constraints are integrated into the graph-based LIO optimization framework.
• Figure 4: Overview of the proposed graph-based localization method using prior topological and metric information. Orange elements form the reference graph, and blue elements form the active graph.
• Figure 5: Visualization of dynamic loading. The prior graph is shown in orange, while the active graph is depicted in blue. For efficiency, only a subset of the geometric model is loaded, specifically the submaps corresponding to vertices within the robot's neighborhood.