Planning and Scheduling in Flexible and Adaptive Microfactories Using Reinforcement Learning

In the context of Industry 4.0, the need for frequent and rapid adaptation of production is becoming increasingly critical. This paradigm shift requires systems capable of autonomously adjusting to dynamic conditions and responding in near real time to the factory’s operational state changes. Existing manufacturers, particularly in precision manufacturing where small batch production is prevalent, face significant difficulties in meeting these requirements. In this paper, we focus on two main challenges: (a) the limited flexibility of production systems, which are often not designed for frequent hardware or software reconfiguration, and (b) complexity in the optimization of production flow, which requires advanced planning and scheduling algorithms. To address these challenges, we propose an innovative method based on reinforcement learning (RL) for planning and scheduling optimization. Reinforcement learning enables systems to adapt efficiently and remain robust to changes in the face of change. We validated our method through simulations conducted on a flexible microfactory, MiLL, developed at Haute Ecole Arc. The RL agent’s performance surpasses a realistic baseline by an average of 8.1%, demonstrating its effectiveness in both planning and scheduling microfactory production. Our method is not confined to precision applications; it can also be applied to any scenario requiring robustness and flexibility.