With the increasing popularity of vertical take-off and landing unmanned aerial vehicles (VTOL UAVs), a new problem arises: pilot training. Most conventional pilot training simulators are designed for full-scale aircrafts, while most UAV simulators are just focused on conceptual testing and design validation. The X-Plane flight simulator was extended to include new functionalities such as complex wind dynamics, ground effect, and accurate real-time weather. A commercial HIL flight controller was coupled with a VTOL convertiplane UAV model to provide realistic flight control. A real flight case scenario was tested in simulation to show the importance of including an accurate wind model. The result is a complete simulation environment that has been successfully deployed for pilot training of the Marvin aircraft manufactured by FuVeX.

Drone navigation is critical, particularly during the initial and final phases, such as the initial ascension, where pilots may fail due to strong external disturbances that could lead to a crash. In this ongoing work, a drone has been successfully trained to perform an ascent of up to 6 meters with external disturbances simulating wind pushing it up to 24 mph, with the DQN algorithm managing external forces affecting the system. It has been demonstrated that the system can control its height, position, and stability in all three axes (roll, pitch, and yaw) throughout the process. The learning process is carried out in the Gazebo simulator, which emulates interferences, while ROS is used to communicate with the agent.

Autonomous unmanned-aircraft obstacle avoidance is a mature research area, but comparatively less work addresses controller-level behavior motivated by European ground-risk and no-overflight considerations during an inspection mission that is already in progress. This paper presents PORCE, a deterministic online local-replanning method for waypoint-preserving obstacle avoidance in autonomous power-line inspection missions. The method integrates a waypoint-driven flight controller, a You Only Look Once-based perception front-end, monocular obstacle geopositioning, and a bounded A-star local planner. Detected protected entities are converted into temporary local exclusion regions; nominal navigation is preserved outside a speed-adaptive reaction horizon and temporarily inhibited when a local conflict is detected. In the logged simulation campaign, the audited encounter triggered at 60.1 m, consistent with the 61 m configured horizon at the 8 m/s cruise speed, and the complete evasion runs added less than 1% path length. Three physical flights over the same corridor produced twenty controlled object encounters; the logged responses included both individual detours and a shared bypass for clustered obstacles, followed by route reattachment. The minimum reported surveyed-footprint clearance was 14.21 m against the 12 m exclusion radius. These results support the feasibility and auditability of the bounded local-replanning architecture within the demonstrated scenarios. The method is regulation-aware by design motivation.