Centralization vs. decentralization in multi-robot sweep coverage with ground robots and UAVs

This dataset accompanies an article and a code repository.In swarm robotics, decentralized control is often proposed as a more scalable and fault-tolerant alternative to centralized control. However, centralized behaviors are often faster and more efficient than their decentralized counterparts. In any given application, the goals and constraints of the task being solved should guide the choice to use centralized control, decentralized control, or a combination of the two. Currently, the exact trade-offs that exist between centralization and decentralization are not well defined. In this paper, we study comparative performance assessment between centralization and decentralization in the example task of sweep coverage, across five different types of multi-robot control structures: random walk, decentralized with beacons, hybrid formation control using self-organizing hierarchy, centralized formation control, and predetermined. In all five approaches, the coverage task is completed by a group of ground robots. In each approach, except for the random walk, the ground robots are assisted by UAVs, acting as supervisors or beacons. We compare the approaches in terms of three performance metrics for which centralized approaches are expected to have an advantage — coverage completeness, coverage uniformity, and sweep completion time — and two metrics for which decentralized approaches are expected to have an advantage — scalability (4, 8, or 16 ground robots) and fault tolerance (0%, 25%, 50%, or 75% ground robot failure). As expected, the results showed that the more centralized approaches greatly outperformed the decentralized ones in terms of coverage completeness, coverage uniformity, and sweep completion time. The decentralized approaches were less affected by robot failures and had better performance gains when the number of robots increased, but unexpectedly, these advantages only made their performance comparable to that of the more centralized approaches, not better than. Finally, we discuss future work on investigating additional conditions (e.g., bottlenecks, supervisor failures, and more complex environments), and on combining the advantages of both centralization and decentralization into one system.