Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

HD-maps as Prior Information for Globally Consistent Mapping in GPS-denied Environments

Waqas Ali, Patric Jensfelt, Thien-Minh Nguyen

TL;DR

The paper tackles GPS-denied localization and long-term mapping by leveraging HD-map priors derived from the drivable-area and ground-height components. It extends a LOAM-based SLAM pipeline with a HD-map constraint module that uses key-frame road points and NDT-based scan-to-map matching to form 2D pose priors, plus altitude from the HD-map, which are integrated into a GTSAM-based pose-graph optimization to achieve a globally consistent 3D map. Evaluations on the Argoverse 2 TbV dataset show significant RMSE reductions and improved robustness over state-of-the-art LOAM variants, demonstrating the practical viability of HD-map priors for GPS-denied operation. The approach also lays groundwork for long-term navigation in familiar environments by enabling compact, map-based updates without external data sources.

Abstract

In recent years, prior maps have become a mainstream tool in autonomous navigation. However, commonly available prior maps are still tailored to control-and-decision tasks, and the use of these maps for localization remains largely unexplored. To bridge this gap, we propose a lidar-based localization and mapping (LOAM) system that can exploit the common HD-maps in autonomous driving scenarios. Specifically, we propose a technique to extract information from the drivable area and ground surface height components of the HD-maps to construct 4DOF pose priors. These pose priors are then further integrated into the pose-graph optimization problem to create a globally consistent 3D map. Experiments show that our scheme can significantly improve the global consistency of the map compared to state-of-the-art lidar-only approaches, proven to be a useful technology to enhance the system's robustness, especially in GPS-denied environment. Moreover, our work also serves as a first step towards long-term navigation of robots in familiar environment, by updating a map. In autonomous driving this could enable updating the HD-maps without sourcing a new from a third party company, which is expensive and introduces delays from change in the world to updated map.

HD-maps as Prior Information for Globally Consistent Mapping in GPS-denied Environments

TL;DR

The paper tackles GPS-denied localization and long-term mapping by leveraging HD-map priors derived from the drivable-area and ground-height components. It extends a LOAM-based SLAM pipeline with a HD-map constraint module that uses key-frame road points and NDT-based scan-to-map matching to form 2D pose priors, plus altitude from the HD-map, which are integrated into a GTSAM-based pose-graph optimization to achieve a globally consistent 3D map. Evaluations on the Argoverse 2 TbV dataset show significant RMSE reductions and improved robustness over state-of-the-art LOAM variants, demonstrating the practical viability of HD-map priors for GPS-denied operation. The approach also lays groundwork for long-term navigation in familiar environments by enabling compact, map-based updates without external data sources.

Abstract

In recent years, prior maps have become a mainstream tool in autonomous navigation. However, commonly available prior maps are still tailored to control-and-decision tasks, and the use of these maps for localization remains largely unexplored. To bridge this gap, we propose a lidar-based localization and mapping (LOAM) system that can exploit the common HD-maps in autonomous driving scenarios. Specifically, we propose a technique to extract information from the drivable area and ground surface height components of the HD-maps to construct 4DOF pose priors. These pose priors are then further integrated into the pose-graph optimization problem to create a globally consistent 3D map. Experiments show that our scheme can significantly improve the global consistency of the map compared to state-of-the-art lidar-only approaches, proven to be a useful technology to enhance the system's robustness, especially in GPS-denied environment. Moreover, our work also serves as a first step towards long-term navigation of robots in familiar environment, by updating a map. In autonomous driving this could enable updating the HD-maps without sourcing a new from a third party company, which is expensive and introduces delays from change in the world to updated map.
Paper Structure (21 sections, 9 equations, 6 figures, 2 tables)

This paper contains 21 sections, 9 equations, 6 figures, 2 tables.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORK
  3. Preliminaries
  4. Notations
  5. Description of HD-map
  6. Problem Statement
  7. Methodology
  8. SLAM
  9. HD-map to pose prior constraint
  10. Pre-Processing
  11. Scan-to-Map Matching
  12. Altitude Constraint Estimation
  13. Pose-Graph Optimization using HD-map Constraints
  14. Experimental Setup
  15. Methods
  16. ...and 6 more sections

Figures (6)

  • Figure 1: Map built from ALOAM aligned with the HD-map along with the ground truth (yellow) and ALOAM (white) trajectories (left) and a more consistent map and the trajectory (red) generated from our system (right).
  • Figure 2: The complete HD-map data of a sequence provided in the Argoverse2-TbV dataset.
  • Figure 3: System overview
  • Figure 4: The distribution of RMSE of ATE erros across the 208 sequences from the Agroverse 2 TbV dataset for different algorithms used in our experiments.
  • Figure 5: RMSE errors of each method for 10 sequences
  • ...and 1 more figures