This thesis explores the use of Reinforcement Learning (RL) for controlling a fixed-wing UAV in attitude and waypoint tracking under varying wind conditions. Unlike traditional controllers such as PID or MPC that rely on accurate models, RL provides a data-driven alternative capable of adapting to nonlinear and uncertain dynamics. Among RL methods, model-based approaches—especially TD-MPC—show strong potential by combining dynamics learning and short-horizon planning. Experiments conducted in JSBSim demonstrate superior tracking performance under nominal wind, though robustness to partially observed disturbances remains challenging. Beyond the experiments, this work contributes a detailed UAV model, a full JSBSim-based simulation environment, and a pedagogical introduction to RL for control. The thesis integrates both attitude and waypoint tracking results to provide a coherent analysis and perspective. Future research envisions hybrid controllers combining data-driven and physics-based models to enhance robustness, interpretability, and reliability of autonomous UAV control.
Federated Learning Semantic Communication in UAV Systems: PPO-Based Joint Trajectory and Resource Allocation Optimization
