A Survey on Global LiDAR Localization: Challenges, Advances and Open Problems

TL;DR

This article aims to provide an overview of recent progress and advancements in LiDAR-based global localization, including recent advancements in several topics, such as maps, descriptor extraction, and cross-robot localization.

Abstract

Knowledge about the own pose is key for all mobile robot applications. Thus pose estimation is part of the core functionalities of mobile robots. Over the last two decades, LiDAR scanners have become the standard sensor for robot localization and mapping. This article aims to provide an overview of recent progress and advancements in LiDAR-based global localization. We begin by formulating the problem and exploring the application scope. We then present a review of the methodology, including recent advancements in several topics, such as maps, descriptor extraction, and cross-robot localization. The contents of the article are organized under three themes. The first theme concerns the combination of global place retrieval and local pose estimation. The second theme is upgrading single-shot measurements to sequential ones for sequential global localization. Finally, the third theme focuses on extending single-robot global localization to cross-robot localization in multi-robot systems. We conclude the survey with a discussion of open challenges and promising directions in global LiDAR localization. To our best knowledge, this is the first comprehensive survey on global LiDAR localization for mobile robots.

Figures (9)

• Figure 1: Fish-shaped paper structure. This survey starts the problem formulation and related introduction at the fish head. Then the fish body part contains the main subtopics of the global LiDAR localization problem: map framework, single-shot and sequential global localization, and cross-robot localization. Finally, an extended discussion on open problems is presented at the fish tail. We also present graphical illustrations above each section title.
• Figure 2: Three typical situations. From top to down: single-robot intra-sequence LCD (loop closing); single-robot inter-sequence re-localization; cross-robot inter-sequence localization. Blue-filled boxes indicate measurements (LiDAR scan or submap). Orange lines are possible relative transformations for global localization problems.
• Figure 3: Four types of single-shot global localization. The term $\mathbf{z}_t$ indicates the input LiDAR point cloud; $\mathbf{M}_{\text{sub}}$ and $\mathbf{M}_{\text{global}}$ represent keyframe-based submaps and a global feature map; place and $\mathbf{m}$ are the retrieved place and submap; $\mathbf{x}_{t}$ is the estimated pose. In Section \ref{['sec:PR']}, place recognition-only approaches provide a retrieved place (keyframe) as the estimated pose. In Section \ref{['sec:PRPE']}, place recognition first provides a prior place then the pose is estimated via an individual pose estimation part. In Section \ref{['sec:PEPR']}, place recognition and pose estimation are coupled together and benefit from shared representations. Methods in Section \ref{['sec:PE']} achieve global pose estimation on a global map where place retrieval is not involved.
• Figure 4: LPD-Net is a place recognition-only approach for global LiDAR localization. Global descriptors are extracted as place descriptions for place retrieval. (Source: LPDNet liu2019lpd, used with permission.)
• Figure 5: (a) Front-end Gaussian modeling and correspondence building. (b) Back-end outlier pruning. The correspondence inlier ratio increases. (c) A challenging global registration result at a road intersection, in which the two LiDAR point clouds are with low overlap and a large view difference. (Source: adapted from G3Reg qiao2023g3reg, used with permission.)