Middle-mile logistics through the lens of goal-conditioned reinforcement learning

2026-05-04

AI Summary

This study addresses the parcel routing optimization problem in medium-haul logistics involving finite-capacity trucks operating across hub-and-spoke networks. It proposes a novel integration of goal-conditioned reinforcement learning with graph neural networks, formulating the task as a multi-objective conditional Markov decision process. The approach leverages graph neural networks to extract compact feature representations from dynamic network states and employs model-free reinforcement learning for efficient decision-making. Compared to existing methods, the proposed framework significantly enhances parcel routing efficiency and transportation resource utilization, enabling effective modeling and optimization of complex, dynamic logistics networks.

Abstract

Middle-mile logistics describes the problem of routing parcels through a network of hubs linked by trucks with finite capacity. We rephrase this as a multi-object goal-conditioned MDP. Our method combines graph neural networks with model-free RL, extracting small feature graphs from the environment state.

Problem

• middle-mile logistics
• goal-conditioned reinforcement learning
• routing
• finite capacity
• hub network

Innovation

• goal-conditioned reinforcement learning
• graph neural networks
• middle-mile logistics
• multi-object MDP
• model-free RL