Recent advancements in the various fields of artificial intelligence have led to many new paradigms to find solutions for engineering challenges that were tackled using classical methods. In this research proposal, we present a new approach for real-time identification of UAV dynamic system parameters. We aim at designing special excitation signals that reveals certain system qualities. The output of the excitation signal is fed into a deep learning classifier that is able to select model parameters that corresponds to the measured system output. The design of the excitation signal can be hardly done using existing analytical methods, instead, a reinforcement learning agent will be trained to select the excitation signal that best suits the selected system model. The presented approach guarantees safety in the identification phase, can be applied in real-time, and results in precise trajectory tracking even in the presence of external disturbances.