Learning-Based Task Assignment for Automated Guided Vehicles: Applying graph neural networks to optimise task assignment in an online warehouse environment

Sammanfattning

Efficient task assignment in multi-Automated Guided Vehicle (AGV) warehouse environmentsis critical for optimizing industrial logistics. In collaboration with MAXAGV,this thesis evaluates the application of reinforcement learning (RL) to addressthis challenge. The warehouse environment is modelled as a graph, and a GraphNeural Network (GNN) policy is trained using Proximal Policy Optimization (PPO)to assign tasks to the vehicle fleet. To capture the complex topology of the facility,which is characterized by long-range spatial configurations and lock-relations thatlimit standard embedding methods like Node2Vec, a novel transductive node embeddingscheme trained via multiple task-specific decoders is introduced. Three coreGNN architectures: Graph Convolutional Networks (GCN), Graph Attention Networks(GAT), and Graph Transformers, along with their heterogeneous extensions,are evaluated and compared against conventional heuristic baselines. The empiricalresults demonstrate the performance trade-offs between the learning-based architecturesand traditional heuristics. Furthermore, the study addresses the broader challengesof deployment, specifically the complexities of reward shaping in real-worldlogistics systems and the systemic barriers to integrating learning-based methodsinto legacy industrial infrastructures.