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

Abstract

Autonomous Aerial Vehicles (AAVs) - known as drones - employment in delivery services is one of the promising transformative technologies. The AAV industry has taken significant steps to develop drones to fulfill the needs of delivery services. However, AAVs have limitations related to the flight range and payload capacity. Therefore, drone route planning is crucial to reducing the effectiveness of these challenges. The recent emergence of Large Language Models (LLMs) has opened new possibilities for solving combinatorial problems using in-context learning (ICL). Unlike traditional machine learning models, LLMs can generate solutions without requiring task-specific fine-tuning by leveraging solved examples within their input prompts. In this study, we explore the application of LLMs to the Drone Routing Problem (DRP), leveraging various ICL strategies to generate optimized delivery routes. Our solution ensures that drone routes are planned to reduce the traveling distance for the full route. Notably, it ensures that drones don't mess any delivery points and fast delivery routes. Through extensive experimentation, we evaluate the effectiveness of different prompt engineering techniques in guiding LLMs to produce high-quality, non-hallucinated route plans. We compared our model results to heuristic-based generated routes to demonstrate the variation between our technique and other techniques. The results demonstrate that LLMs, when properly prompted, can reliably generate valid routing solutions, highlighting their potential as a flexible and adaptive tool for drone logistics planning. Project link: https://github.com/ahmed-abdulhuy/Solve-TSP-using-GPT3.5.git.