General Place Recognition Survey: Towards Real-World Autonomy

TL;DR

This survey frames place recognition (PR) as a foundational component for real-world, long-term robotic autonomy and situates it within the SLAM2.0 paradigm. It analyzes definitions (position-based vs overlap-based), formalizes an effective PR criterion, and catalogs low-level and high-level place representations, emphasizing multi-modal and cross-domain embeddings, graphs, and foundation-model–driven approaches. It discusses core challenges—appearance and viewpoint variations, generalization, efficiency, and uncertainty—and surveys solutions spanning place modeling, sequence-based matching, geometric and hybrid methods, plus strategies for lifelong learning and efficiency. The paper also surveys applications (long-term navigation, visual terrain relative navigation, multi-agent localization, lifelong autonomy) and provides public datasets and evaluation tools, thereby offering a comprehensive framework and resources to advance PR toward real-world deployment and cross-domain robotics.

Abstract

In the realm of robotics, the quest for achieving real-world autonomy, capable of executing large-scale and long-term operations, has positioned place recognition (PR) as a cornerstone technology. Despite the PR community's remarkable strides over the past two decades, garnering attention from fields like computer vision and robotics, the development of PR methods that sufficiently support real-world robotic systems remains a challenge. This paper aims to bridge this gap by highlighting the crucial role of PR within the framework of Simultaneous Localization and Mapping (SLAM) 2.0. This new phase in robotic navigation calls for scalable, adaptable, and efficient PR solutions by integrating advanced artificial intelligence (AI) technologies. For this goal, we provide a comprehensive review of the current state-of-the-art (SOTA) advancements in PR, alongside the remaining challenges, and underscore its broad applications in robotics. This paper begins with an exploration of PR's formulation and key research challenges. We extensively review literature, focusing on related methods on place representation and solutions to various PR challenges. Applications showcasing PR's potential in robotics, key PR datasets, and open-source libraries are discussed. We conclude with a discussion on PR's future directions and provide a summary of the literature covered at: https://github.com/MetaSLAM/GPRS.

Figures (12)

• Figure 1: Structure of our General Place Recognition (PR) Survey. PR is the ability to recognize visited areas under different environmental conditions and viewpoint differences. This survey is structured as follows: Section \ref{['sec:definition_challenges']} defines the problem of position-based PR and introduces the significant challenges. Section \ref{['sec:representation']} investigates methods in place representation. Section \ref{['sec:solution']} and Section \ref{['sec:application']} provide the solutions for the current four major challenges and potential applications, respectively, Finally, Section \ref{['sec:data_eval']} introduces the current datasets, metrics, and related supported libraries for PR research.
• Figure 2: This timeline maps the evolution of PR from handcrafted to data-driven methods, analyzing key techniques, surveys, and applications. Our survey emerges at an opportune moment, given major events across multiple fields including embodied AI, reconstruction, and collaborative perception. Citations for select works are omitted due to limited page; readers may search referenced titles online.
• Figure 3: In position-based PR, the focus is on identifying if a query image stays at the same location as a database image. For instance, in the provided images, position-based PR would recognize the Query $1$ image as matching the database image, but reject the Query $2$ that is taken from a geographically distant location and considered as a different place. Overlap-based PR, however, would classify both query images as the same place since they share visual overlap with the database, denoted by the red box. But the Query $2$ offers limited utility for downstream navigation tasks.
• Figure 4: Challenges in real-world PR. In real-world navigation tasks, robots may encounter the following challenges: (a) changing visual appearances due to temporal variations (lighting, seasons) burnett2023boreas, (b) diverse viewpoint differences for the same areas, (c) visiting new unknown areas warburg2020mapillary, (d) impacts to efficiency when deployed on real-world robots yin2023bioslam, and (e) uncertainty estimation of data and modal dolezal2022uncertainty.
• Figure 5: Diverse Sensor Modalities and Observation Properties. The top box contains various camera setups with different lens scaramuzza2014omnidirectional and imaging sensors jiao2022fusionportable. The bottom-left box shows major LiDAR types, point cloud, and multi-channel images using the cylinder projection from the point cloud chen2021overlapnet. The bottom-right box shows a typical RaDAR and data represented in polar and cartesian images hong2022radarslam.