Deep reinforcement learning for dynamic scheduling of energy-efficient automated guided vehicles

Automated guided vehicle (AGV) scheduling has become a hot topic in recent years as manufacturing systems become flexible and intelligent. However, little research regards dynamic AGV scheduling considering energy consumption, particularly battery replacement. This paper proposes a novel method that employs deep reinforcement learning to address the dynamic scheduling of energy-efficient AGVs with battery replacement in production logistics systems. The bi-objective joint optimization problem of AGV scheduling and battery replacement management is modeled as a Markov Decision Process, which supports data-driven decision-making. Then, this paper constructs a deep reinforcement learning-based optimization architecture and develops a novel dueling deep double Q network algorithm to maximize the long-term rewards for optimizing material handling’s tardiness and energy consumption. Numerical experiments and a case study demonstrate that the proposed algorithm is more efficient and cleaner than state-of-the-art methods. The proposed method can significantly improve customer satisfaction and reduce production costs within flexible manufacturing processes, particularly in Industry 4.0.