Employing in-context learning prompts with large language models for drone routing in delivery services

Area of Science:

• Logistics and Operations Research
• Artificial Intelligence
• Robotics

Background:

• Autonomous Aerial Vehicles (AAVs), or drones, present a transformative technology for delivery services.
• Current AAVs face significant limitations in flight range and payload capacity, necessitating efficient route planning.
• The Drone Routing Problem (DRP) is a critical challenge in optimizing AAV delivery operations.

Purpose of the Study:

• To investigate the application of Large Language Models (LLMs) for solving the Drone Routing Problem (DRP).
• To leverage In-Context Learning (ICL) strategies within LLMs to generate optimized delivery routes.
• To evaluate the effectiveness of prompt engineering techniques for generating high-quality, non-hallucinated drone routes.

Main Methods:

• Utilized LLMs with various In-Context Learning (ICL) strategies to address the DRP.
• Employed prompt engineering techniques to guide LLM-generated route plans.
• Compared LLM-generated routes against heuristic-based methods for performance evaluation.

Main Results:

• LLMs, when appropriately prompted, can reliably generate valid and optimized drone routing solutions.
• The study demonstrated the potential of LLMs to reduce traveling distance and ensure all delivery points are met.
• LLM-generated routes showed variations compared to traditional heuristic-based approaches.

Conclusions:

• LLMs show significant promise as a flexible and adaptive tool for drone logistics and route planning.
• The findings highlight the capability of LLMs to address complex combinatorial problems like the DRP without task-specific fine-tuning.
• Further research into prompt engineering can enhance LLM performance in autonomous delivery systems.