Autonomous systems have been widely analyzed in (structured) environments, such as industrial robotics, household automation, and autonomous driving. Cooperative autonomous systems in predefined environments are also an ongoing topic of research. However, the ability of such systems to generalize and adapt in environments with frequent and significant changes, like the SFG, remains largely underexplored. But exactly this capability is crucial considering the dynamic nature of distributed production environments such as the SFG. There challenges like rapidly shifting operational configurations or high variability in task demands arise, necessitating the need for robots featuring enhanced reasoning and robustness.

This research combines theory, experiments, and simulations: It begins with a literature survey on the "Sense-Plan-Act" cycle and End-to-End approaches in single and multi-agent robotics, focusing on techniques for cooperative data collection and task execution. Emphasis is placed on real-time data sharing and decision-making in dynamic environments.

Additionally, an experimental setup will feature various robots each equipped with sensors, compute, and actuators, enabling testing of cooperative perception strategies and algorithms for resource allocation in the real world. Paralleled by this, a simulated environment using tools like NVIDIA Omniverse shall enable scalable testing and development, while also allowing for research into bridging the sim-to-real gaps.