Model-Based Thermal Updraft State Estimation for Fixed-Wing UAVs Using Aerodynamic Moments and Reinforcement Learning

Using thermal updrafts can significantly improve the endurance of unmanned aerial vehicles (UAVs). This paper proposes a model-based approach for enhancing thermal exploitation, where aerodynamic roll and pitch moments induced by thermals are integrated as observations into a reinforcement learning (RL) framework. Simulation results show that incorporating both roll and pitch moments provides richer environmental cues, enabling the RL agent to more accurately locate thermal cores and improve climbing efficiency. Compared to relying solely on vertical velocity measurements, the proposed method achieves faster convergence and higher precision in thermal exploitation. This work offers valuable insights for developing intelligent, energy-efficient soaring strategies in longendurance UAV applications.