A Model-Based Formalization for the Development of Information Processing Units through Deep Reinforcement Learning

Abstract

In today’s landscape of increasingly complex systems, designing a suitable information processing unit presents a paramount challenge for developers. This paper introduces a novel model-based formalization using the Systems Modeling Language (SysML) for the development of information processing units through deep reinforcement learning (DRL). The approach at hand offers a systematic framework that harnesses the benefits of modularity, formalization, and traceability that SysML offers with the capabilities of DRL to adapt to diverse and various tasks, with the use of virtual simulations to address design challenges. By integrating the strengths of SysML with DRL, the method enables developers to adapt their solutions to dynamic changes in system requirements, task descriptions, and operational environments. The modularity of the proposed approach, coupled with the use of virtual simulations, allows for targeted and sustainable adjustments, diminishing development time, costs, and turnaround. To validate the practicality of the proposed method, the authors present two case studies: one for initial development and another specifically focused on change management within the domain of autonomous robotics. Experimental results underscore the aforementioned benefits of the approach and its adaptability to dynamic changes. This approach shows considerable promise for enhancing model-based, DRL-driven decision-making capabilities across a range of engineering applications.