Flexible robotic cell scheduling with graph neural network based deep reinforcement learning

Flexible robotic cells are pivotal in flexible and customized manufacturing. An effective scheduling policy for such cells can significantly reduce the makespan and improve the production efficiency. This study introduces an innovative end-to-end real-time scheduling method leveraging deep reinforcement learning (DRL) to minimize the makespan in a flexible robotic cell. We introduce a heterogeneous disjunctive graph model for a nuanced representation of the scheduling problem, which incorporates transportation through specific disjunctive arcs. The DRL utilizes Graph Neural Network (GNN) for model feature extraction and employs Proximal Policy Optimization (PPO) to train the scheduling agent. Our methodology can also better leverage the transport robot capacity to mitigate system blockage and deadlock. Numerical experiments are conducted to demonstrate the effectiveness of the proposed method.