This paper presents a generalized approach for Imitation Learning (IL) control, employing least-squares optimization to replicate an expert behavior for a given task. While Nonlinear Model Predictive Control (NMPC) is renowned for its predictive capabilities, constraint-handling and adaptability, its computational complexity often inhibits real-time implementation. IL control overcomes this challenge by establishing a data-driven input-output mapping function represented in quadratic forms, enabling it to effectively mimic NMPC's performance. The effectiveness of this approach is demonstrated through simulations conducted on a fixed-wing F-18 aircraft with 6 degrees-of-freedom rigid body dynamics. The study highlights the potential of least-squares optimization techniques and quadratic functions in IL. By leveraging NMPC-generated data, IL approach replicates complex behaviors in real-time while minimizing control signal fluctuations, promising advancements in autonomous flight in dynamic aerospace environments.
Real-Time Imitation NMPC Using Least-Squares Optimization for Aircraft Trajectory Tracking
29.09.2024
865795 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
An improved NMPC-NDOB scheme for trajectory tracking of unmanned surface vessel
SAGE Publications | 2022
|NMPC trajectory planner for urban autonomous driving
Taylor & Francis Verlag | 2023
|A Comparative Study between NMPC and Baseline Feedback Controllers for UAV Trajectory Tracking
DOAJ | 2023
|NMPC and genetic algorithm based approach for trajectory tracking and collision avoidance of UAVs
BASE | 2013
|