Solving pickup and drop-off problem using hybrid pointer networks with deep reinforcement learning.

Area of Science:

• Operations Research
• Artificial Intelligence
• Computer Science

Background:

• The Pickup and Drop-off Problem (PDP) is a complex logistical challenge with significant real-world applications.
• Existing methods often struggle with capacity constraints and intricate node relationships.
• Optimizing tour length is crucial for efficiency in logistics and transportation.

Purpose of the Study:

• To develop a general and effective method for solving the Pickup and Drop-off Problem (PDP).
• To reduce the overall tour length traveled by an agent.
• To ensure adherence to truck capacity restrictions and node-to-node relationships.

Main Methods:

• Utilizing Hybrid Pointer Networks (HPNs) as the primary machine learning model.
• Integrating Deep Reinforcement Learning (DRL) to optimize decision-making.
• Addressing two demand patterns: balanced (sum to zero) and unbalanced (practical scenarios).

Main Results:

• Achieved state-of-the-art results compared to existing models for the PDP.
• Demonstrated the effectiveness of the proposed HPNs and DRL approach.
• Successfully handled complex node interactions and capacity constraints.

Conclusions:

• The proposed HPNs and DRL method offers a powerful solution for the PDP.
• The approach is effective for both balanced and unbalanced demand scenarios.
• Publicly available data, models, and code facilitate further research and application.