CNN based precise nonlinear tracking control for a nano unmanned helicopter: Theory and implementation

This paper proposes a deep convolutional neural network (CNN)-based geometric integral control strategy for a nano unmanned helicopter, which weighs less than 70 g and has a fuselage length of less than 0.25 m. Compared to existing nonlinear controllers, the proposed method offers significant advantages in terms of feasibility, easy tuning of parameters, and data requirements. The deep CNN-based system identification effectively captures the complex dynamics of the nano helicopter, enabling accurate modeling and compensation for uncertainties. The geometric integral control strategy enhances the system's robustness against unknown external disturbances, and ensures stable and precise flight performance. The feasibility of the proposed method is demonstrated through real-time flight experiments, which show strong robustness and accurate trajectory tracking performance. Additionally, the tuning process is simplified due to the adaption nature of the deep learning-based approach, reducing the need for extensive parameter adjustments. The data requirements are also minimized, as the deep CNN can be trained with a relatively small dataset, making the method more practical for real-world applications. The results indicate that the proposed method outperforms traditional control strategies, particularly in terms of handling modeling uncertainties and external disturbances.