EdgeSLAM 2.0: Enhancing Scalability in Multi-Agent Systems

In the chapter, we delve into how SwarmMap (a.k.a., EdgeSLAM 2.0) revolutionizes collaborative visual SLAM services in edge computing environments, specifically tailored for burgeoning applications like search-and-rescue operations, inventory management automation, and industrial inspections. Central to SwarmMap’s architecture are three innovative modules: a server–client synchronization process based on change logs, a scheduler attuned to task priority, and a streamlined global map representation. Together, they efficiently mitigate issues like data overflow, excessive bandwidth use, and inaccuracies in localization, which intensify with an increase in agent numbers. SwarmMap is made ROS-compatible, and its resources have been made publicly available (Access the code and resources at https://github.com/MobiSense/SwarmMap .). This enables existing visual SLAM solutions to readily adopt SwarmMap, significantly boosting their multi-agent handling capabilities. Through rigorous testing and a detailed 3-month deployment in one of the globe’s most extensive oil-fields, SwarmMap demonstrated its ability to support double the agents ( \(\ge \) 20) compared to leading methods, while ensuring resource utilization remains constant. Additionally, it achieves a remarkable average trajectory accuracy of 38 cm, surpassing prior achievements by more than 55%.