Towards a Hybrid Architecture for Self-Adaptive and Self-Organizing Systems

In multi-agent systems, control can either be initiated from a central coordination unit, or each agent decides on its own about its next adaptation step to adjust to changes in the dynamic environment. To counteract a single point of failure as well as egoistic adaptation decisions, we propose the establishment of a hybrid architecture. Starting from static, fully centralized coordination and static, fully decentralized decision-making, the systems become adaptive through optimization techniques and reinforcement learning, respectively, before being merged into the hybrid system. To evaluate and compare the systems' performance, the coordination of platooning, where vehicles drive with a low inter-vehicle distance to reduce drag and improve traffic flow, serves as a running example.