Compound Learning-Based Model Predictive Control Approach for Ducted-Fan Aerial Vehicles.

Area of Science:

• Robotics and Control Systems
• Aerospace Engineering
• Artificial Intelligence

Background:

• Designing effective control systems for ducted-fan unmanned aerial vehicles (DFUAVs) is challenging due to complex dynamics and aerodynamic configurations.
• Existing control methods often rely on known physics or are limited in scope, necessitating new approaches for uncertain systems.

Purpose of the Study:

• To propose a compound learning-based model predictive control (MPC) framework for DFUAVs.
• To develop a system with efficient dynamics learning and robust disturbance rejection capabilities.
• To ensure stability and recursive feasibility for the proposed control scheme.

Main Methods:

• Offline sparse identification to obtain a nominal model of the DFUAV.
• Online reinforcement learning (RL) to generate initial control input sequences.
• MPC-driven optimization that updates the learned model with real system data for improved efficiency.

Main Results:

• The proposed compound learning-based MPC framework demonstrates efficient dynamics learning for DFUAVs.
• The approach achieves effective disturbance rejection, enhancing control performance.
• Computational efficiency is improved through online model updates within the MPC framework.

Conclusions:

• The developed compound learning-based MPC approach offers a viable solution for controlling DFUAVs with uncertain dynamics.
• The framework successfully integrates sparse identification and reinforcement learning for enhanced control.
• The study validates the efficacy and stability of the proposed control scheme through comparative analysis.