The City that Never Settles: Simulation-based LiDAR Dataset for Long-Term Place Recognition Under Extreme Structural Changes

TL;DR

The paper tackles long-term place recognition in outdoor urban environments undergoing extreme structural changes due to construction and demolition. It introduces CNS, a synthetic LiDAR dataset generated with the CARLA simulator, together with a symmetric temporal-change metric, $TCR_{sym}$, to quantify changes independent of source–target ordering. The authors evaluate four LiDAR-PR baselines and show substantial performance degradation as changes intensify, demonstrating that existing methods struggle with city-scale transformations. By providing a reproducible benchmark with quantified change metrics, this work enables development of more robust long-term PR methods for real-world construction and demolition scenarios.

Abstract

Large-scale construction and demolition significantly challenge long-term place recognition (PR) by drastically reshaping urban and suburban environments. Existing datasets predominantly reflect limited or indoor-focused changes, failing to adequately represent extensive outdoor transformations. To bridge this gap, we introduce the City that Never Settles (CNS) dataset, a simulation-based dataset created using the CARLA simulator, capturing major structural changes-such as building construction and demolition-across diverse maps and sequences. Additionally, we propose TCR_sym, a symmetric version of the original TCR metric, enabling consistent measurement of structural changes irrespective of source-target ordering. Quantitative comparisons demonstrate that CNS encompasses more extensive transformations than current real-world benchmarks. Evaluations of state-of-the-art LiDAR-based PR methods on CNS reveal substantial performance degradation, underscoring the need for robust algorithms capable of handling significant environmental changes. Our dataset is available at https://github.com/Hyunho111/CNS_dataset.

Figures (5)

• Figure 1: System Configuration
• Figure 2: The upper displays aerial views with trajectories overlaid for Riverside 02, Suburbia 02, Downtown 04, and Metropolis 04. The below shows all the routes on each map, with red being the starting point and blue being the ending point.
• Figure 3: Visualization of the set $\mathcal{H}$ forMetropolis. The left two columns illustrate how $\mathcal{H}$ changes under the source–target ordering, causing variations in TCR values. To ensure consistency regardless of the ordering, we introduce the union of both sets, as shown in the right column, which remains invariant to the ordering.
• Figure 4: AUC for each baseline as time intervals increase
• Figure 5: Precision–Recall curves of each baseline for Metropolis