LEMON-Mapping: Loop-Enhanced Large-Scale Multi-Session Point Cloud Merging and Optimization for Globally Consistent Mapping

TL;DR

This work re-examine the role of loops for multi-robot mapping and introduces spatial bundle adjustment for multi-robot maps, reducing divergence and eliminating blurring in overlaps and designs a PGO-based approach that leverages refined bundle adjustment constraints to propagate local accuracy to the entire map.

Abstract

Multi-robot collaboration is becoming increasingly critical and presents significant challenges in modern robotics, especially for building a globally consistent, accurate map. Traditional multi-robot pose graph optimization (PGO) methods ensure basic global consistency but ignore the geometric structure of the map, and only use loop closures as constraints between pose nodes, leading to divergence and blurring in overlapping regions. To address this issue, we propose LEMON-Mapping, a loop-enhanced framework for large-scale, multi-session point cloud fusion and optimization. We re-examine the role of loops for multi-robot mapping and introduce three key innovations. First, we develop a robust loop processing mechanism that rejects outliers and a loop recall strategy to recover mistakenly removed but valid loops. Second, we introduce spatial bundle adjustment for multi-robot maps, reducing divergence and eliminating blurring in overlaps. Third, we design a PGO-based approach that leverages refined bundle adjustment constraints to propagate local accuracy to the entire map. We validate LEMON-Mapping on several public datasets and a self-collected dataset. The experimental results show superior mapping accuracy and global consistency of our framework compared to traditional merging methods. Scalability experiments also demonstrate its strong capability to handle scenarios involving numerous robots.

Figures (17)

• Figure 1: The merging map of our framework in island sequence of MARS-LVIG lvig dataset, the details in the figure are framed and shown in two forms: side view (SV) and bird’s eye view (BEV).
• Figure 2: The framework of our method. Our framework takes multi-robot submaps ,odometry and loop closures as input, and generates accurate and globally consistent merged map. (a) The Loop Processing Module, which removes outliers and recall false negative loops, providing correct and sufficient loops for the spatial BA. (b) The Map Merging Module, which achieves accurate and globally consistent multi-robot map merging through two-step PGO and spatial BA. Each of the three steps in map merging module interacts closely with the multi-robot pose graph.
• Figure 3: An example of loop classification in S3E Campus_1 dataset s3e, the trajectories of multi-robot and the two types of loops are shown in the figure.
• Figure 4: The red and green nodes in the left part show the trajectories of different robots at certain loop. (a) and (b) show the submaps before BA optimization and after BA optimization respectively. Our spatial bundle adjustment significantly reduces divergence and generates locally consistent map.
• Figure 5: (a) shows the multi-robot poses in a spatial window. (b) shows the reordered poses using PCA for spatial HBA.

Theorems & Definitions (9)

• Lemma 1: Continuity in Eigenvalue Functions
• proof
• Lemma 2: Computational Complexity Reduction of DBA
• proof
• Lemma 3: Covariance ordering: DBA vs joint BA
• proof
• proof
• proof
• proof