Meta-path-based Graph Reinforcement Learning Framework for Flexible Job Shop Scheduling Problem with Limited Mobile Robots

As mobile robots (MRs) are increasingly integrated into flexible manufacturing systems (FMS), transportation has emerged as a pivotal component that significantly impacts production throughput and scheduling effectiveness. To tackle this issue, this paper introduces a meta-path-based graph reinforcement learning (MP-GRL) methodology for the flexible job shop scheduling problem with limited mobile robots (FJSP-LMRs), with the objective of makespan minimization. First, the decision-making procedure of FJSP-LMRs is modeled as a Markov decision process (MDP). Second, scheduling states are represented using a heterogeneous graph, and a meta-path-based graph neural network (MP-GNN) is utilized for the extraction of structural information. Third, the proximal policy optimization (PPO) algorithm is applied to optimize the policy network. Comprehensive experimental results demonstrate the effectiveness of the developed MP-GRL framework.