Optimizing a Dynamic Vehicle Routing Problem with Deep Reinforcement Learning: Analyzing State Space Components

This paper develops a reinforcement learning approach for solving dynamic vehicle routing problems in real-time. The study systematically analyzes which state space components are most critical for effective RL-based routing decisions.

Key Contributions

• RL-based algorithm for dynamic vehicle routing with real-time adaptability
• Systematic ablation studies on state space component importance
• 15% efficiency improvement over greedy heuristics
• Improved robustness under operational disruptions

Methodology

Multi-agent RL approach combined with optimization heuristics, with detailed analysis of how different state representations affect solution quality and convergence behavior.

Recommended citation: Konovalenko, A., & Hvattum, L.M. (2023). "Optimizing a Dynamic Vehicle Routing Problem with Deep Reinforcement Learning: Analyzing State Space Components." Logistics.
Download Paper