A Digital Twin-Level Modeling Framework for UAV Control Algorithm Optimization and Sim2real

As multicopter unmanned aerial vehicles (UAVs) are increasingly adopted in industrial and manufacturing applications, there is a growing demand for robust, efficient, and cost-effective control algorithm development. However, traditional trial-and-error approaches relying heavily on real-flight experiments remain time-consuming and resource-intensive, particularly in dynamic or uncertain environments. To address these challenges, this paper presents a digital twin-level modeling framework that enables highfidelity simulation, accelerated controller design, and seamless sim2real transfer. The framework integrates detailed physical modeling of multicopter dynamics, realistic sensor emulation, and hardware-in-the-loop (HIL) simulation within a modular architecture. Each subsystem—including propulsion, aerodynamics, ground interaction, and onboard sensors—is systematically constructed and verified using experimental data. A representative application is demonstrated by designing a model-based linear quadratic regulator (LQR) controller trained using hybrid data composed of digital simulations and a small amount of real-world flight data. The controller is evaluated under external wind disturbances and benchmarked against baseline PX4 and hand-tuned PID controllers. Quantitative comparisons using standard performance metrics demonstrate that the hybrid-trained LQR outperforms alternatives while significantly reducing tuning time and testing costs. The results confirm the proposed digital twin framework as a promising tool for enhancing the efficiency, reliability, and scalability of UAV control algorithm development.