Multi-Task Deep Reinforcement Learning for Intelligent Logistics Path Planning and Scheduling Optimization

Abstract

Intelligent logistics systems optimize path planning and scheduling problems by introducing deep reinforcement learning (DRL) to cope with the complexity of dynamic demand and resource constraints. This study proposes an improved strategy that combines multi-task learning (MTL) with Q-learning to achieve simultaneous optimization of multiple related subtasks, such as path selection, task scheduling, and resource allocation, and shares some network parameters to improve model efficiency and generalization ability. The experimental design covers a variety of logistics scenarios, including urban distribution, long-distance transportation, and emergency response. Five baseline models are used for comparative evaluation to verify the advantages of the DRL method in computational efficiency, cost control, resource utilization, service level, and dynamic adaptability. Through experimental verification, the DRL model achieved a 15% cost reduction in cost control compared to traditional algorithms; the resource utilization rate reached 85%, and it performed excellently in terms of efficiency improvement. It has excellent adaptability when dealing with dynamic demands and can respond quickly to environmental changes, effectively improving the overall performance of the intelligent logistics system. It is particularly suitable for application scenarios that require real-time decision-making and support dynamic demand fluctuations.