Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GroundLoc: Efficient Large-Scale Outdoor LiDAR-Only Localization

Nicolai Steinke, Daniel Goehring

TL;DR

GroundLoc tackles large-scale outdoor LiDAR-only localization by projecting LiDAR data into 3-channel BEV images of the ground and matching against compact prior maps. It supports either a learned keypoint framework (R2D2) or a classical descriptor (SIFT) for a 3-DOF pose estimate, using an approximate KD-tree matcher and a Quatro-based estimator with damped online corrections. The prior maps are lightweight GeoTIFF rasters (~4 MB/km^2) generated from a single drive, enabling fast, sensor-agnostic deployment. Across SemanticKITTI and HeLiPR, GroundLoc delivers state-of-the-art localization accuracy with online runtimes above 14 Hz and robust performance in single- and multi-session scenarios, and the authors provide open-source code to foster broader adoption.

Abstract

In this letter, we introduce GroundLoc, a LiDAR-only localization pipeline designed to localize a mobile robot in large-scale outdoor environments using prior maps. GroundLoc employs a Bird's-Eye View (BEV) image projection focusing on the perceived ground area and utilizes the place recognition network R2D2, or alternatively, the non-learning approach Scale-Invariant Feature Transform (SIFT), to identify and select keypoints for BEV image map registration. Our results demonstrate that GroundLoc outperforms state-of-the-art methods on the SemanticKITTI and HeLiPR datasets across various sensors. In the multi-session localization evaluation, GroundLoc reaches an Average Trajectory Error (ATE) well below 50 cm on all Ouster OS2 128 sequences while meeting online runtime requirements. The system supports various sensor models, as evidenced by evaluations conducted with Velodyne HDL-64E, Ouster OS2 128, Aeva Aeries II, and Livox Avia sensors. The prior maps are stored as 2D raster image maps, which can be created from a single drive and require only 4 MB of storage per square kilometer. The source code is available at https://github.com/dcmlr/groundloc.

GroundLoc: Efficient Large-Scale Outdoor LiDAR-Only Localization

TL;DR

GroundLoc tackles large-scale outdoor LiDAR-only localization by projecting LiDAR data into 3-channel BEV images of the ground and matching against compact prior maps. It supports either a learned keypoint framework (R2D2) or a classical descriptor (SIFT) for a 3-DOF pose estimate, using an approximate KD-tree matcher and a Quatro-based estimator with damped online corrections. The prior maps are lightweight GeoTIFF rasters (~4 MB/km^2) generated from a single drive, enabling fast, sensor-agnostic deployment. Across SemanticKITTI and HeLiPR, GroundLoc delivers state-of-the-art localization accuracy with online runtimes above 14 Hz and robust performance in single- and multi-session scenarios, and the authors provide open-source code to foster broader adoption.

Abstract

In this letter, we introduce GroundLoc, a LiDAR-only localization pipeline designed to localize a mobile robot in large-scale outdoor environments using prior maps. GroundLoc employs a Bird's-Eye View (BEV) image projection focusing on the perceived ground area and utilizes the place recognition network R2D2, or alternatively, the non-learning approach Scale-Invariant Feature Transform (SIFT), to identify and select keypoints for BEV image map registration. Our results demonstrate that GroundLoc outperforms state-of-the-art methods on the SemanticKITTI and HeLiPR datasets across various sensors. In the multi-session localization evaluation, GroundLoc reaches an Average Trajectory Error (ATE) well below 50 cm on all Ouster OS2 128 sequences while meeting online runtime requirements. The system supports various sensor models, as evidenced by evaluations conducted with Velodyne HDL-64E, Ouster OS2 128, Aeva Aeries II, and Livox Avia sensors. The prior maps are stored as 2D raster image maps, which can be created from a single drive and require only 4 MB of storage per square kilometer. The source code is available at https://github.com/dcmlr/groundloc.

Paper Structure

This paper contains 17 sections, 7 equations, 3 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. GroundLoc Localization Pipeline
  4. Creation of the LiDAR BEV Images
  5. Map Creation
  6. BEV Images Keypoint and Descriptor Extraction
  7. Training of R2D2
  8. Calculating the Position Estimate
  9. Experiments and Evaluation
  10. Parameters
  11. Metrics
  12. Matching Evaluation
  13. Single Session Localization
  14. Multi-Session Localization
  15. Ablation Studies
  16. ...and 2 more sections

Figures (3)

  • Figure 1: Visualization of multi-session localization results on the HeLiPR Roundabout Ouster sequence of our method. The coloring of the trajectory indicates the translational deviation from the ground truth.
  • Figure 2: Overview of the proposed system. Input point cloud in red, intermediate results in blue, and output pose estimate in green.
  • Figure 3: Visualization of the BEV images: a) BEV intensity image with SIFT keypoints; b) 3-channel BEV image with R2D2 keypoints; c) Repeatability map (low-to-high colormap from red to green); d) Reliability map (low-to-high colormap from red to green).