Reinforcement learning in dynamic job shop scheduling: a comprehensive review of AI-driven approaches in modern manufacturing

Abstract Dynamic job shop scheduling (DJSS) demands real-time adaptability under unpredictable conditions such as sudden job arrivals, equipment failures, and fluctuating demands. Traditional scheduling approaches—though foundational—often fall short when faced with rapid changes and high computational complexity. Recent developments in artificial intelligence (AI), especially reinforcement learning (RL), offer powerful alternatives by continuously refining scheduling policies through interaction with live shop-floor data. This review systematically examines AI-driven scheduling methods, highlighting how evolutionary heuristics, advanced machine learning, and RL-based algorithms each address the demands of modern manufacturing. Emphasis is placed on RL’s capacity to cope with large state spaces, handle continuous or discrete control, and integrate domain heuristics for more robust real-time decision-making. Despite these advances, challenges remain in algorithm scalability, interpretability, data availability, and standardization of performance metrics. Future directions point toward leveraging digital twins, quantum computing, hybrid models, and explainable RL to ensure more resilient, transparent, and scalable solutions. By illuminating both current achievements and persistent gaps, this review underscores the transformative potential of RL in dynamic scheduling and highlights actionable steps for broader industrial adoption.