MLATC: Fast Hierarchical Topological Mapping from 3D LiDAR Point Clouds Based on Adaptive Resonance Theory

TL;DR

The paper presents MLATC, a hierarchical extension of ART-based ATC-DT for fast global topological mapping from sequential 3D LiDAR point clouds. By organizing nodes across multiple resolutions with adaptive layer addition and a top-down nearest-neighbor search, MLATC achieves sublinear search costs and millisecond per-frame runtimes in large-scale, dynamic environments. The approach preserves the stability-plasticity balance of ART while enabling scalable, real-time map building, as demonstrated on synthetic and campus-scale real-world LiDAR data. These results indicate MLATC's potential for enabling real-time, structure-aware navigation in large, unknown environments without extensive hyperparameter tuning.

Abstract

This paper addresses the problem of building global topological maps from 3D LiDAR point clouds for autonomous mobile robots operating in large-scale, dynamic, and unknown environments. Adaptive Resonance Theory-based Topological Clustering with Different Topologies (ATC-DT) builds global topological maps represented as graphs while mitigating catastrophic forgetting during sequential processing. However, its winner selection mechanism relies on an exhaustive nearest-neighbor search over all existing nodes, leading to scalability limitations as the map grows. To address this challenge, we propose a hierarchical extension called Multi-Layer ATC (MLATC). MLATC organizes nodes into a hierarchy, enabling the nearest-neighbor search to proceed from coarse to fine resolutions, thereby drastically reducing the number of distance evaluations per query. The number of layers is not fixed in advance. MLATC employs an adaptive layer addition mechanism that automatically deepens the hierarchy when lower layers become saturated, keeping the number of user-defined hyperparameters low. Simulation experiments on synthetic large-scale environments show that MLATC accelerates topological map building compared to the original ATC-DT and exhibits a sublinear, approximately logarithmic scaling of search time with respect to the number of nodes. Experiments on campus-scale real-world LiDAR datasets confirm that MLATC maintains a millisecond-level per-frame runtime and enables real-time global topological map building in large-scale environments, significantly outperforming the original ATC-DT in terms of computational efficiency.

Figures (8)

• Figure 1: Hierarchical structure of the proposed MLATC. The first layer $G^{(1)}$ is a fine-resolution topological map directly learned from LiDAR point clouds. Upper layers $G^{(2)}, \dots, G^{(L)}$ summarize disjoint child node sets into coarser-resolution nodes, and the top layer maintains a single root node. Parent--child links between layers are used to constrain the search space during hierarchical nearest-neighbor search.
• Figure 2: Processing time comparison on synthetic data ($N\!\le\!10^4$).
• Figure 3: Scalability of MLATC on synthetic data up to $N=10^7$. The network formed up to 11 active layers during the sequence.
• Figure 4: Aerial view of the real-world experiment. Red: LiDAR trajectory; blue arrow: reference point A. Basemap: aerial photograph by Geospatial Information Authority of Japan (GSI) gsi2025, edited to overlay the trajectory and marker.
• Figure 5: Per-frame processing time on real-world LiDAR data (small-scale regime).