Eco-SLAM: Resource-Efficient Edge-Assisted Collaborative Visual SLAM System

Collaboration among multiple smart agents such as robots and UAVs is critical for the key tasks of simultaneous localization and mapping (SLAM), which are essential for many robotics applications. Visual SLAM maps by multiple collaborative agents can be promptly generated with the assistance of edge servers in proximity as the computing infrastructures. However, as the number of agents connected to an edge server continues growing, the pressure on bandwidth consumption and resource utilization also climbs dramatically, which may trigger the failure of map generation. To tackle these challenges, this article proposes Eco-SLAM, a resource-efficient edge-assisted collaborative multi-agent visual SLAM system. Eco-SLAM has been designed to enable large-scale parallelism in SLAM framework and optimized for use in both edge servers and intelligent agents. The unique Core-tr and CoMap library design ensures efficient utilization of resources and consistent data. Additionally, Eco-SLAM incorporates the ORB-based image compression algorithm, which optimizes data transmission with constrained networking resources. We implement and evaluate the Eco-SLAM system in a real environment and demonstrate its effectiveness through extensive experiments ranging from the public SLAM datasets to realistic deployment scenarios. Thorough evaluations show that Eco-SLAM can reduce the memory consumption of other multi-agent SLAM frameworks by up to 20.1% during runtime on the edge server, and save up to 25.1% wireless bandwidth consumption without compromising the accuracy of the map generation.