Quadcopter Trajectory Generation Based on Large Language Model

Quadcopter trajectory generation is a critical aspect of autonomous drone navigation, particularly in dynamic and complex environments. Traditional methods such as Rapidly-exploring Random Trees (RRT) and A* algorithms, often require extensive manual programming and graphical user interfaces, which limits their accessibility for non-technical users. In this paper, we propose a novel approach leveraging Large Language Models (LLMs) for intuitive and flexible trajectory generation using natural language commands. The proposed system integrates LLMs with the Robot Operating System 2 (ROS 2) framework to translate user commands into executable flight paths. The model employs meta-prompting techniques to enhance command interpretation accuracy and structure the generated trajectories in JSON format for real-time execution. Experimental validation in a simulated environment demonstrates that the proposed approach achieves a 92.3 success rate across various trajectory scenarios, including circular, rectangular, and triangular paths. Furthermore, error analysis based on Mean Absolute Error (MAE) and Integral Absolute Error (IAE) confirms the model's stability and accuracy in maintaining the intended flight path. These findings highlight the potential of LLM-based trajectory generation to improve user interaction, reduce complexity, and enhance the autonomy of quadcopter navigation systems. © 2025 IEEE.