On Reinforcement Learning for Part Dispatching in UAV-Served Flexible Manufacturing Systems

The industrial environment of the past years has been characterized by a high rate of change, pushing the industry to implement innovative technologies to satisfy market needs. Unmanned aerial vehicles (UAVs) and Reinforcement Learning (RL) are being implemented in the manufacturing industry to meet changing market demands for efficiency. This work focuses on using RL for optimal part dispatching in Flexible Manufacturing Systems (FMS) using UAVs. A virtual discrete events model is used to represent the shop floor state and a reward function is defined to maximize production. Proximal Policy Optimization (PPO) is employed to train the RL agent. Results show a production increase of up to 145.16% compared to traditional heuristic rules.