Distributed control and organization of communicating mobile robots : design, simulation, and experimentation

Abstract

Leveraging the communication-based cooperation of multiple robotic systems has the potential to significantly further the state of the art of what is achievable with robotic automation. Therefore, beyond solely improving the capabilities of individual robotic agents, reconfigurable robotic networks have come to the attention of research and industry. However, despite the potential to increase flexibility, robustness, and performance, robotic networks are not yet in widespread application, with many research challenges remaining. After all, developing a reliable, reconfigurable distributed system is very difficult, adding to the manifold, interdisciplinary challenges posed by robotics in general. Hence, to better understand and subsequently overcome these challenges, distributed robotics is still in a state where it can benefit significantly from research that tackles well-defined benchmark problems. Consequently, this thesis faces the challenges of distributed robotics at the example of a cooperative transportation task. In the task, omnidirectional mobile robots cooperate to maneuver polygonal objects purely by pushing forces and in a completely self-reliant manner. The task is found to be a formidable benchmark problem since it raises all major challenges of the field while still being easily graspable, intuitively making evident the qualities of the control and organization schemes employed. The thesis discusses all aspects of the task in an encompassing manner, not only including the design of the employed control and organization methods, but also the software architecture and even the custom robotic hardware employed. Results from simulations and real-world hardware experiments show that the proposed scheme is of unprecedented versatility, putting into practice all major promises of distributed robotics, including plug-and-play control for online reconfigurations of the robotic network. This is achieved by relying, at heart, on optimization-based schemes. The task is decomposed into a formation control task and the organizational task of inferring formations useful for manipulation, allowing the usage of distributed model predictive control for formation control and of distributed optimization for organization. Further challenges dealt with include self-reliant task allocation as well as local and global navigation. However, the contributions of the thesis extend beyond the immediate needs of the benchmark problem. A component that may prove helpful in other research endeavors in the field includes the devised distributed software architecture, which greatly facilitates the transition from simulations to experiments. Similarly, the custom mobile robot and different proposed setups of the formation controller are also suited to other tasks and projects. Due to formation control’s universal appeal, the proposed approach based on distributed predictive control is analyzed separately from the transportation task. In experiments, the predictive control-based approach confirms its theory-rooted advantages in comparison to a more traditional approach, despite the latter being modified to also respect input constraints. Finally, a proposed distributed version of an augmented Lagrangian particle swarm optimization algorithm, which is used to devise formations in the thesis, may even prove useful far beyond robotics.