Landing a spacecraft is a challenging task as there are different external (environment) and internal (measurements) factors that play an essential role in precisely landing the ship in the target with small error margins. Even though the task is difficult, the success of such projects can lead to large cost reductions through energy and resource efficiency. Model Predictive Control can help to achieve the task but requires reasonably accurate models of the real spacecraft. The task is non-trivial as the dynamics of the spacecraft are nonlinear and uncertain, which leads to difficulty in constructing models. Different Machine Learning techniques could be used to build such models, however, this results in a nonlinear complex representation that is difficult to understand. The processes and outcomes of such models are difficult to interpret, and they usually function as a ”black box” mapping of input and output. Thus, the main idea of this thesis is to use the Koopman operator to model the nonlinear dynamics of the spacecraft into a higher dimensional lifted feature space where its evolution is linearly represented. The linear operators built on the Koopman operator framework are simple, data-driven, and used to design controllers, in our case linear Model Predictive Control without any nonlinear optimization schemes. Furthermore, we explore how the predictors obtained through this method are comparatively superior in terms of performance to the existing linear predictors. When used in a Model Predictive Control scheme, the model is efficient, allowing for quick control loop optimization. As a result of the effective linear model controlling techniques, it is suitable for real-time systems. Landing a spaceship has many requirements that are similar to the stabilization of an inverted pendulum: the spaceship is controlled from below through the thrusters, which act as the cart in the pendulum. The process of building predictors and applying Model Predictive Control is similar. That is why a pendulum system is used as an illustrative example in this study.     «

Landing a spacecraft is a challenging task as there are different external (environment) and internal (measurements) factors that play an essential role in precisely landing the ship in the target with small error margins. Even though the task is difficult, the success of such projects can lead to large cost reductions through energy and resource efficiency. Model Predictive Control can help to achieve the task but requires reasonably accurate models of the real spacecraft. The task is non-trivi...     »