Adaptive and Learning-based NMPC Strategies for Quadrotor Control

Thanks to hardware and software evolution, machine learning implementations have become more viable. Nowadays the possibility to handle precise data acquisition and high computational capabilities is available also for embedded devices used in drones. This thesis aims at implementing and comparing the results of an Adaptive Non- linear Model Predictive Controller (Adaptive+NMPC) and two Learning-based NMPC (Lb-NMPC) methods, in a trajectory tracking task for a quadrotor, where different types of disturbances and model mismatches need to be rejected.

Thanks to hardware and software evolution, machine learning implementations have become more viable. Nowadays the possibility to handle precise data acquisition and high computational capabilities is available also for embedded devices used in drones. This thesis aims at implementing and comparing the results of an Adaptive Non- linear Model Predictive Controller (Adaptive+NMPC) and two Learning-based NMPC (Lb-NMPC) methods, in a trajectory tracking task for a quadrotor, where different types of disturbances and model mismatches need to be rejected. The Lb-NMPC approaches are respectively based on Gaussian Processes and Neural Networks.