Light-weight collaborative mapping via three-dimensional semantic descriptor matrix

• The collaborative localization and mapping of multiple robots have significant implications in unknown environments. However, it faces challenges such as poor communication among team members and inefficiency in global map construction. Therefore, this work proposes a collaborative visual SLAM framework supporting semantic mapping in large-scale environments using three-dimensional semantic descriptors.This system consists of two parts: the robot side and the server side. The former includes image tracking and semantic segmentation, while the latter involves semantic object representation and global semantic map generation.The proposed three-dimensional semantic descriptors elevate the multi-robot SLAM system beyond constructing semantic point cloud maps, enriching map representation by describing multi-level, multi-category semantic objects.Finally, the system is extensively evaluated on benchmark datasets and compared with other state-of-the-art multi-robot SLAM systems, demonstrating lower communication requirements and multi-level semantic map representation. • This article analyzes the front-end and back-end tasks of the multi-robot SLAM system, completes task separation and unloads cloud computing capabilities. It combines deep learning semantic segmentation technology with requirements such as communication bandwidth, front-end system stability, and system real-time performance to design an SLAM system. A three-dimensional semantic descriptor that does not depend on image information but only relies on two attributes, namely three-dimensional spatial coordinates and semantic labels, is proposed, along with a multi-map matching algorithm based on this descriptor. The system achieves precise fitting of regular and irregular objects and ultimately constructs multi-level, multi-type semantic maps through multi-robot collaboration, enriching the representation form and expanding the functionality of the map. • We deeply appreciate your consideration of our manuscript, and we look forward to receiving comments from reviewers. If you have any queries, please contact me at the address below. Multi-robot collaborative Simultaneous Localization and Mapping (SLAM) faces increasing challenges in communication, computation, and data storage. To address these issues, this paper proposes a novel SLAM algorithm based on a collaborative 3D feature semantic description matrix. First, a multi-agent collaborative mapping framework and data interaction model are developed, followed by a semantic map fusion mechanism that clusters point clouds into semantic nodes to enrich structural representation. Based on these semantic nodes and their spatial neighborhoods, a 3D feature semantic description matrix is constructed to jointly encode geometric attributes and local three-dimensional topological structures, enabling efficient and accurate map matching and fusion. Experimental results on the KITTI dataset show that the proposed method reduces computational complexity by approximately 63%, increases map fusion speed by about 62%, and decreases communication data volume by around 84% (≈ 6 × compression), while maintaining registration accuracy within 1.2% of the baseline. These results demonstrate that the proposed approach effectively improves computational and communication efficiency while preserving accuracy, providing a promising solution for large-scale multi-robot collaborative SLAM. The reconstructed maps and trajectories derived by the framework with four robots. Re-constructed map and camera trajectory.