Imitation Learning (IL) has emerged as a powerful technique for replicating expert behavior, offering reduced computational demands and simplified programming. However, traditional IL approaches often lack interpretability, transparency, and stability guarantees, factors that are critical for safety and reliability in high-stakes applications. This dissertation addresses these challenges by proposing a simple yet robust IL-driven framework that emphasizes minimalistic training, ease of implementation, and broad generalization across complex, real-time control tasks. At the core of this research is the construction of a quadratic policy that effectively captures the relationship between input states and output actions, an essential requirement for high-performing IL systems. By utilizing linear least-squares regression to derive a quadratic mapping function, the IL framework enhances transparency and adaptability, overcoming the limitations of conventional black-box methods. Furthermore, by leveraging Nonlinear Model Predictive Control (NMPC) as the expert, the proposed approach inherits the optimality and precision of NMPC while achieving real-time performance through data-driven policy learning. The principal contributions of this dissertation include: (1) development of a generalized, statistical-based IL framework applicable to multiple vehicle types and control missions; (2) introduction of a modular IL approach for efficient learning in high-dimensional systems with low-resource training; (3) use of NMPC as an expert for path-following in dynamic environments; (4) establishment of theoretical asymptotic stability guarantees via Lyapunov-based analysis; (5) robustness validation under disturbances such as wind and sensor noise; and (6) demonstration of cross-platform adaptation on selected vehicles. Through extensive simulation and rigorous evaluation, the imitation learning-based framework demonstrates strong performance, adaptability, and practical feasibility. This research strengthens the foundations of imitation learning by bridging the gap between model-based control and learning-based methods. It offers a stable, scalable, and interpretable solution for real-time control, enabling automated systems to operate safely and effectively in dynamic environments.     «

Imitation Learning (IL) has emerged as a powerful technique for replicating expert behavior, offering reduced computational demands and simplified programming. However, traditional IL approaches often lack interpretability, transparency, and stability guarantees, factors that are critical for safety and reliability in high-stakes applications. This dissertation addresses these challenges by proposing a simple yet robust IL-driven framework that emphasizes minimalistic training, ease of implement...     »