Requirements for Generating Learning Environments for Autonomous Systems Behavior in a Digital Continuum

Autonomous systems in material handling are increasingly prevalent in logistics, offering benefits such as flexibility, adaptability, robustness, and sustainability. To fully harness these advantages, a novel paradigm, the Digital Continuum, is proposed for the development and operation of such systems. A critical component of the Digital Continuum is a deeply integrated digital system model, which serves as a simulation, training, and test environment for virtual agents
corresponding to physical robots. To ensure robust performance in learned behavior, a large number of learning environments is needed, thus highlighting the importance of an automated generation process. This process can significantly reduce modeling effort and is yet to be developed. This paper presents the
derivation of requirements for an automated learning environment generation approach, unifying elements from Digital Continua, intralogistics, and robotics domains. Furthermore, the paper briefly discusses the research gap in the context of existing procedural content generation and domain randomization
approaches. By addressing these requirements and bridging the research gap, a generation approach has the potential to profoundly facilitate the development and operation of autonomous systems in logistics.