PRISM-Loc: a Lightweight Long-range LiDAR Localization in Urban Environments with Topological Maps

TL;DR

PRISM-Loc addresses city-scale outdoor localization by combining a compact topological map with a curb-aware LiDAR scan-matching refinement that operates directly on raw scans. The method uses MinkLoc3D for fast place recognition to identify candidate locations, followed by a curb-enhanced 2D scan-matching stage to precisely align scans within a selected topological node, enabling lightweight, real-time localization on embedded platforms. Experiments on ITLP-Campus and Oxford RobotCar show robust localization with mean ATE around 0.5 m and modest map sizes (~20 MB), outperforming memory-heavy baselines and offering strong scalability for urban robotics. The contributions include a compact city-scale map representation, a novel curb-detection plus scan-matching pipeline, and a comprehensive evaluation highlighting efficiency and robustness under challenging urban conditions.

Abstract

We propose PRISM-Loc - a lightweight and robust approach for localization in large outdoor environments that combines a compact topological representation with a novel scan-matching and curb-detection module operating on raw LiDAR scans. The method is designed for resource-constrained platforms and emphasizes real-time performance and resilience to common urban sensing challenges. It provides accurate localization in compact topological maps using global place recognition and an original scan matching technique. Experiments on standard benchmarks and on an embedded platform demonstrate the effectiveness of our approach. Our method achieves a 99\% success rate on the large-scale ITLP-Campus dataset while running at 150 ms per localization and using a 20 MB map for localization. We highlight three main contributions: (1) a compact representation for city-scale localization; (2) a novel curb detection and scan matching pipeline operating directly on raw LiDAR points; (3) a thorough evaluation of our method with performance analysis.

Figures (7)

• Figure 1: Ground truth metric map (top), metric map built by RTAB-Map (middle), and topological map aligned with a ground-truth metric map (bottom).
• Figure 2: An example of a topological map built by PRISM-TopoMap method (red nodes and blue edges), and a 2D grid of a location.
• Figure 3: A scheme of the proposed global localization pipeline. It includes $F_{PR}$ place encoder (we use MinkLoc3D) and the proposed scan matching module. The output of the pipeline is the current location and the robot's relative pose in this location.
• Figure 4: Scan matching comparison: our previous algorithm (left) muravyev2025prism and the proposed algorithm (right). The candidate scan is shown in gray and white, the reference scan is shown in green (wall points) and red (curb points)
• Figure 5: Curb detection output. Points belonging to curbs are shown in red.