Abstract

We propose LLM-DaaS, a novel Drone-as-a-Service (DaaS) framework thatleverages Large Language Models (LLMs) to transform free-text user requestsinto structured, actionable DaaS operation tasks. Our approach addresses thekey challenge of interpreting and structuring natural language input toautomate drone service operations under uncertain conditions. The system iscomposed of three main components: free-text request processing, structuredrequest generation, and dynamic DaaS selection and composition. First, wefine-tune different LLM models such as Phi-3.5, LLaMA-3.2 7b and Gemma 2b on adataset of text user requests mapped to structured DaaS requests. Usersinteract with our model in a free conversational style, discussing packagedelivery requests, while the fine-tuned LLM extracts DaaS metadata such asdelivery time, source and destination locations, and package weight. The DaaSservice selection model is designed to select the best available drone capableof delivering the requested package from the delivery point to the nearestoptimal destination. Additionally, the DaaS composition model composes aservice from a set of the best available drones to deliver the package from thesource to the final destination. Second, the system integrates real-timeweather data to optimize drone route planning and scheduling, ensuring safe andefficient operations. Simulations demonstrate the system's ability tosignificantly improve task accuracy, operational efficiency, and establishLLM-DaaS as a robust solution for DaaS operations in uncertain environments.