Three-dimensional (3D) point cloud map merging is a pivotal technology in robotics and automation, enabling the integration of multiple 3D point cloud maps into a single, comprehensive representation of the environment. This technique is particularly advantageous in multi-robot coordination, where multiple robots collaborate to explore and map extensive areas. Each robot generates a local map within its local frame, which serves as crucial data for localization, collision avoidance, navigation, and path planning, and can later be shared and fused into a global map. In addition, human operators in the industry can find this advantageous, as it allows for faster and more efficient inspections without the need for manual map alignment. This thesis introduces a modular framework for autonomous 3D point cloud map merging in multi-robot systems, addressing the challenge of aligning local maps by identifying acceptable spatial coordinate transformations. This framework facilitates real-time map merging during multi-robot exploration, enhancing mapping efficiency by preventing redundant exploration of already mapped areas. The first contribution stems from formulating and addressing the map merging problem through a modular pipeline and evaluating each component. Then two methods are presented that improve place recognition performance, a fundamental aspect of the process. The first method extends the place recognition pipeline with a topological classification module, enhancing performance in challenging environments and autonomously triggering the map merging pipeline for higher success rates. The second method integrates additional data modalities, such as an inexpensive Wi-Fi module, to enhance place recognition performance. Furthermore, the thesis addresses communication challenges in multi-robot systems. A solution for centralized systems is proposed, where a control mechanism regulates map data transmission to ensure critical information is preserved and the map merging process is not compromised. Additionally, a combined solution for place recognition descriptors is presented, which compresses LiDAR data to improve transmission efficiency. Finally, the map merging framework serves as the backbone of a change detection algorithm. Both the map merging framework and the change detection algorithm are evaluated through a series of use-case deployments, including autonomous Unmanned Aerial Vehicles (UAVs) operations in mining areas and a safety inspection mission following a real blast.