In the rapidly evolving domain of unmanned aerial vehicle (UAV) navigation, the accurate prediction of UAV trajectories presents significant challenges, particularly given the inadequacies of comprehensive datasets suitable for training robust AI models. Traditional trajectory prediction methods based on constant velocity or acceleration, while maturing in 2D domains such as autonomous vehicles, fail to address the complex, high-speed dynamics characteristic of multi-rotor UAVs operating in 3D space. To bridge this gap, we introduce a comprehensive, open-source simulation framework that integrates Robot Operating System (ROS 2), the Gazebo robotics simulator, and PX4 open-source autopilot. Our framework is not restricted to fixed data outputs; rather, it is highly modular and capable of generating an arbitrary number of trajectories with user-defined characteristics. As a significant contribution, we provide an example dataset encompassing over 5000 random UAV trajectories generated across 20 h of simulated flight. This framework excels at generating realistic UAV trajectory simulations and preparing extensive, high-quality synthetic datasets. Unlike existing approaches such as Flightmare and MidAir, our solution supports robust preprocessing for dataset readiness, including uniform resampling, statistical normalization, and trajectory segment generation, crucial for effective AI training. We also provide our open-source implementation to help users customize their datasets.