Online self-learning attitude tracking control of morphing unmanned aerial vehicle based on dual heuristic dynamic programming

This article presents a partially model-free dual heuristic dynamic programming (DHP) control approach to address the online discrete-time attitude tracking problem. The DHP method learns the control policy by solving the translated optimal regulation problem of the augmented system, which is constructed with the tracking error dynamics and the desired trajectory. In the context of the augmented system, a corresponding recursive least squares (RLS) identification strategy has been devised, and the DHP update rules have been derived. The identified augmented system model not only provides dynamic information and predicted states for the actor and critic updates but also incorporates an uncertainty compensation part to reduce the learning complexity and enhance robustness of the controller. It is noteworthy that concurrent learning (CL) is employed to improve the critic convergence by utilizing an identified model-based data generation method. This approach eliminates the need for an offline learning stage and enables flexible controller design based on the available system knowledge. Based on the proposed approach, different attitude control frameworks have been designed for a morphing unmanned aerial vehicle (UAV). In addition to numerical simulations, flight tests have been conducted to further demonstrate the capabilities of the proposed approach, highlighting its effectiveness and advantages.