Digital twin-driven deep reinforcement learning for real-time optimisation in dynamic AGV systems
Donggun Lee, Yong-Shin Kang, Sang Do Noh
Abstract
Automated guided vehicle (AGV) systems play a critical role in production logistics and workflow management within manufacturing. As manufacturing environments become increasingly dynamic, real-time optimisation of AGV routing and path planning has become essential. Many studies have applied deep reinforcement learning (DRL) to address these challenges. However, DRL often lacks adaptability to real-time environmental changes. To overcome this limitation, digital twin (DT) technologies have been explored alongside DRL. Most existing studies, however, utilise DT only as a post-training validation tool. This study proposes a novel DT-driven DRL approach in which DT actively participates in both training and decision-making phases. A real-time optimisation system is designed for dynamic AGV operations, and its effectiveness is validated through an industrial case study. Results demonstrate the proposed approach’s capability to significantly enhance adaptability and efficiency in complex manufacturing settings.