Development of Optimal Control Strategies For Quadrotor Force Compensation and Trajectory Tracking

Modern industrial automation requires Unmanned Aerial Vehicles (UAVs) to become active physical agents capable of exerting controlled forces on their surroundings, a domain known as Aerial Physical Interaction (APhI). This thesis addresses the automation of a manual buffer management operation, where a constant resistive force must be applied continuously to the advancing leading edge of a compliant product. The proposed solution employs an autonomous quadrotor equipped with a rigid probe that must simultaneously track a sweeping trajectory and regulate contact force. An optimal control pipeline is developed and verified across a dual environment methodology: a Python digital twin and a Software-in-the-Loop (SITL) architecture integrating ROS 2, Gazebo and PX4, in which the custom controllers inject wrench setpoints directly into the control allocator. A central contribution is the comparative evaluation between a Linear Quadratic Regulator with integral action (LQRI) and a Nonlinear Model Predictive Controller (NMPC). Both controllers track feedforward references generated via differential flatness. Their comparison shows comparable tracking performance: the LQRI compensates force indirectly through integral action, whereas the NMPC regulates a predicted contact force explicitly, holding the highest force target more tightly with a smoother actuator signal. Both controllers are robust to moderate model errors, whereas the stock PX4 cascaded controller cannot hold the contact force. Further tests show that the main limitation of planar quadrotors in this task is yaw authority, which saturates under lateral contact moments regardless of the control algorithm used. Cross-validation shows that the Python digital twin is a faithful predictor of the realistic experiment and acts as a lower bound on its tracking error. Future developments include a canted-rotor airframe for direct yaw control and a GPS-denied indoor localization pipeline.