OSM-SLAM: Aiding SLAM with OpenStreetMaps priors

Matteo Frosi, Veronica Gobbi, Matteo MatteucciFrontiers Robotics AI 2023 · DOI · OpenReview

Abstract

Traditional Simultaneous Localization and Mapping (SLAM) systems typically operate without prior environmental knowledge, building maps from scratch. This paper introduces OSM-SLAM, a framework that integrates 2D building footprints from OpenStreetMap (OSM) as spatial priors for LiDAR-based SLAM. By aligning real-time 3D sensor data with global map data, the system improves localization accuracy, enables map correction, and provides a mechanism for re-localization after sensor failure. Evaluations on the KITTI dataset demonstrate significant reductions in trajectory error and enhanced robustness in urban environments.

Introduction

Most state-of-the-art SLAM systems follow a "tabula rasa" approach, relying solely on onboard sensors to navigate unknown environments. While effective, this ignores the vast amount of existing global spatial data. Furthermore, traditional SLAM struggles with re-localization when sensors fail or tracking is lost. OSM-SLAM addresses these issues by leveraging freely available OSM data as a "global anchor," allowing the robot to utilize prior knowledge while remaining robust to inaccuracies in the public map.

Methods

OSM-SLAM is built upon the ART-SLAM framework, a modular LiDAR-based Graph SLAM system. The architecture integrates a dual-branch pipeline to merge 3D LiDAR data with 2D map priors.

System architecture of OSM-SLAM, highlighting the integration of OSM Map Builder and Pose Graph Builder modules.

Pipeline Overview

The system processes data through two primary branches:

  1. Point Cloud Processing: Vertical planes (walls) are extracted from 3D LiDAR scans using a modified RANSAC algorithm and projected onto a 2D plane.
  2. OSM Data Integration: Building footprints are queried via the Overpass API based on the robot's estimated position and converted into a 2D point cloud.
Detailed pipeline for point cloud processing, OSM building map creation, and data association via ICP.

Optimization Strategies

Data association is performed using Iterative Closest Point (ICP) to align detected walls with OSM footprints. These matches form constraints in a pose graph optimized via the g2o framework. The authors propose three strategies:

  • Prior SLAM: Treats OSM buildings as fixed references.
  • Rigid SLAM: Allows buildings to move as a single block to fit sensor data.
  • Non-rigid SLAM: Treats buildings as independent entities, allowing the robot to correct displacements in the OSM map.

Results

The system was evaluated using the KITTI Odometry Dataset (Sequences 00 and 07). In scenarios where standard loop closure was disabled, Prior SLAM reduced the Mean Absolute Trajectory Error (ATE) on Sequence 07 from 1.537m to 1.166m. The system also demonstrated superior stability and successful re-localization after intentional tracking losses of 100 frames, a scenario where baseline systems typically fail. Qualitatively, the Non-rigid SLAM approach successfully "nudged" OSM footprints to match real-world observations, effectively refining the public map.

Conclusion

OSM-SLAM demonstrates the value of integrating global map priors into local SLAM pipelines. While the current implementation faces computational bottlenecks in plane extraction and depends on building-rich environments, it provides a robust foundation for "map-aware" robotics. Future work will explore deep learning for segmentation and the inclusion of additional OSM features like road networks and poles.

Summary updated

Abstract

p>In the last decades, Simultaneous Localization and Mapping (SLAM) proved to be a fundamental topic in the field of robotics, due to the many applications, ranging from autonomous driving to 3D reconstruction. Many systems have been proposed in literature exploiting a heterogeneous variety of sensors. State-of-the-art methods build their own map from scratch, using only data coming from the equipment of the robot, and not exploiting possible reconstructions of the environment. Moreover, temporary loss of data proves to be a challenge for SLAM systems, as it demands efficient re-localization to continue the localization process. In this paper, we present a SLAM system that exploits additional information coming from mapping services like OpenStreetMaps, hence the name OSM-SLAM, to face these issues. We extend an existing LiDAR-based Graph SLAM system, ART-SLAM, making it able to integrate the 2D geometry of buildings in the trajectory estimation process, by matching a prior OpenStreetMaps map with a single LiDAR scan. Each estimated pose of the robot is then associated with all buildings surrounding it. This association allows to improve localization accuracy, but also to adjust possible mistakes in the prior map. The pose estimates coming from SLAM are then jointly optimized with the constraints associated with the various OSM buildings, which can assume one of the following types: Buildings are always fixed (Prior SLAM); buildings surrounding a robot are movable in chunks, for every scan (Rigid SLAM); and every single building is free to move independently from the others (Non-rigid SLAM). Lastly, OSM maps can also be used to re-localize the robot when sensor data is lost. We compare the accuracy of the proposed system with existing methods for LiDAR-based SLAM, including the baseline, also providing a visual inspection of the results. The comparison is made by evaluating the estimated trajectory displacement using the KITTI odometry dataset. Moreover, the experimental campaign, along with an ablation study on the re-localization capabilities of the proposed system and its accuracy in loop detection-denied scenarios, allow a discussion about how the quality of prior maps influences the SLAM procedure, which may lead to worse estimates than the baseline.</p>
BibTeX
@article{frosi2023osm,
  author = {Matteo Frosi and Veronica Gobbi and Matteo Matteucci},
  title = {{OSM-SLAM: Aiding SLAM with OpenStreetMaps priors}},
  journal = {Frontiers Robotics AI},
  year = {2023},
  doi = {10.3389/frobt.2023.1064934},
}
Figure