This dissertation develops numerical optimization methods for quadratic programming problems whose structural properties challenge solver reliability and computational efficiency. These methods are applied within optimal control problem formulations, with particular emphasis on autonomous unmanned aerial vehicles (UAVs). To support a practical deployment, a modular optimal control framework is developed and integrated with communication protocols and simulation environments. A stabilized primal-dual interior-point solver for quadratic programming (QP) is presented, integrating proximal and augmented Lagrangian regularization to enhance both robustness and convergence. A novel proximal regularization for slack variables is proposed, along with safeguards to maintain trajectories near the central-path. Extensive numerical experiments on challenging benchmark problems, including ill-conditioned and poorly scaled cases, demonstrate the solver’s accuracy, robustness, and competitiveness with leading commercial and open-source QP solvers. Moreover, a reinforcement learning approach is introduced to automatically tune proximal and augmented Lagrangian regularization weight dynamics. The policy, trained on a small number of random QP instances, adapts the solver parameters online, enhancing convergence speed while maintaining theoretical guarantees. The QP solver is further embedded into an in-house Sequential Quadratic Programming (SQP) framework, enabling the efficient solution of constrained nonlinear optimization problems and providing comprehensive background for the following part of the thesis. A Model Predictive Control (MPC) framework is developed and applied to three types of vehicles: fixed-wing aircraft, quadrocopters, and ground vehicles, each presenting different dynamics and control challenges. The methods are validated through Software-In-The-Loop (SITL) simulations and real-world experiments, including flight tests with a very-light aircraft. A real-time trajectory visualization and guidance interface was implemented to assist pilots, demonstrating the practical relevance of the approach in both civil and military air traffic scenarios. Finally, interactions between cooperative autonomous systems are examined through the high-speed landing of a quadrocopter on an unmanned ground vehicle, demonstrating the ability of MPC-based methods to handle multi-agent coordination, high-speed maneuvers, and strict feasibility requirements. Overall, the dissertation shows that advanced optimization techniques, combined with data-driven parameter tuning, can provide reliable and efficient control of autonomous systems in complex, real-world environments.     «

This dissertation develops numerical optimization methods for quadratic programming problems whose structural properties challenge solver reliability and computational efficiency. These methods are applied within optimal control problem formulations, with particular emphasis on autonomous unmanned aerial vehicles (UAVs). To support a practical deployment, a modular optimal control framework is developed and integrated with communication protocols and simulation environments. A stabilized primal-...     »