Abstract

Robots are leaving the friendly, well-structured world of automation and are facing the challenges of a dynamic world. The uncertain conditions in the dynamic world call for a high degree of robustness and adaptivity for individual robots as well as interactions between multiple robots and other system entities. The uncertainty makes it difficult for designers and engineers to anticipate all conditions, interactions, and side effects a system will have to deal with while the system is specified and developed. Furthermore, implicit and explicit coordination is required to perform a joint goal with multiple entities in a multi-robot system. Enabling scalability for multi-robot applications can be especially supported by means of implicit and decentralised coordination approaches. Nevertheless, robots that adapt to the dynamic environment and coordinate themselves still have to pursue their given tasks or goals. This thesis researches how multi-purpose, mobile, multi-robot systems can be enhanced to operate more adaptively in dynamic environments. This is done by analysing and exploring the combination of so far separated research directions of goal-driven decision-making and planning as well as self-adaptation and self-organisation. The presented hybrid decision-making and planning framework is integrated into the popular robotic middleware \gls{ROS}. The solution combines symbolic planning with reactive behaviour networks, automated task delegation, reinforcement learning, and pattern-based selection of suitable self-organisation mechanisms. On that account, it brings together the advantages of bottom-up and top-down oriented architectures for task-level control of multi-robot systems. The developed framework enables a coherent and integrated design and implementation of decision-making and planning as well as coordination application logic within one software ecosystem that features a common domain model and a modular architecture. This results in a simplification of the development of adaptive multi-purpose multi-robot systems by avoiding system discontinuities and enabling a holistic view on the actual implementation. The presented approach has been successfully evaluated in various research projects and international competitions in the field of robotics and multi-agent systems.

@PhdThesis{hrabia2019thesis,
Title = {{Self-adaptive and self-organised planning and decision-making for multi-robot systems}},
Author = {Christopher-Eyk Hrabia},
School = {Technische Universit\"at Berlin},
Year = {2019},
Doi = {10.14279/DEPOSITONCE-9337}
}