Digital Twin-Driven Reinforcement Learning for Dynamic Path Planning of AGV Systems

With the digitalization of industries, manufacturing systems are becoming increasingly complex and diverse. There is a growing focus on production flexibility and automation, leading to active research on production logistics (PL) systems that can effectively address these challenges. PL systems significantly influences the quality and productivity of products, and the proper design and optimization of path planning for logistics robots are crucial. This paper proposes a novel approach, Digital Twin (DT)-driven reinforcement learning (RL) for dynamic path planning of automated guided vehicles (AGVs) systems. The complex real-world path planning problem is represented as a Markov Decision Process (MDP) and a DT-based Q learning algorithm that can solve the represented path planning problem is proposed. To validate the effectiveness and adaptability of the proposed approach, a system is implemented and applied to an actual manufacturing site.