Reinforcement Learning Lifecycle for the Design of Advanced Robotic Systems

摘要

Machine learning is a recognised technology for problem solving and accelerates the automation by enabling systems to act independently. Cyber-physical systems based on machine learning enable factories to increase the skills of robotic systems. The lack of standardised tools and workflows for artificial intelligence (AI) rises the importance of research methodologies and frameworks for industrial application. First concepts have shown potential for a combination of holistic development methodologies and AI. We present a Reinforcement Learning Lifecycle (RLL) for the development of advanced robots. The autonomous software agent can furthermore lead to the automated optimisation of the systems. The virtual-based method speeds up learning processes, improve development and operation processes by the evaluation of multiple simulation environments. We show how an AI-based methodology assists the development of advanced robots along the product lifecycle. The first implementations show potential regarding the usability and results of the approach during different development processes.