Multi-aerial rotor vehicles, such as quadcopters and hexacopters, are increasingly finding application in confined environments, ranging from indoor surveillance and inspection to warehouse automation. Achieving precise and agile control in these constrained spaces presents unique challenges, which include navigating limited space, coping with intricate environmental dynamics, the aerodynamic disturbances of the confined spaces and real-time decision-making constraints. Data-driven modelling incorporated Nonlinear Model Predictive Control (NMPC) is proposed as a suitable solution for addressing these challenges, with a focus on learning for improved adaptability and the capacity to model and adapt to disturbances in confined space environments.
This thesis explores if the integration of data-driven modelling and model-based controllers can improve the performance of Multi-Rotor Aerial Vehicles(MRAV) in confined environments. The potential combination of data-driven modelling and NMPC control offers several advantages in the context of confined environment applications. The major advantage is that the data-driven model provides a more accurate representation of the disturbances, allowing for improved tracking and robustness in the face of external disturbances which is crucial when navigating around obstacles.
Firstly, the dynamic effects of flying an MRAV in a confined space are analysed based on measurements collected from physical experiments. Then, an NMPC controller that incorporates a Gaussian Process model is explored, proposed, implemented, analysed, and validated with real-time simulations. The proposed controller is compared with a nominal NMPC and NMPC controller that incorporates a state-of-the-art Disturbance Observer. The results show that the proposed controller outperforms both the nominal NMPC controller and the NMPC controller with the Disturbance Observer.