Output feedback control of a quadrotor UAV using neural networks

Area of Science:

• Robotics
• Control Systems Engineering
• Artificial Intelligence

Background:

• Quadrotor unmanned aerial vehicles (UAVs) require sophisticated control strategies, especially in outdoor environments where dynamics are unpredictable.
• Traditional control methods often rely on accurate models of UAV dynamics, which are difficult to obtain in real-world scenarios.
• Underactuation in quadrotors limits control over all six degrees of freedom (DOF) using standard four-input configurations.

Purpose of the Study:

• To propose a novel nonlinear output feedback controller for quadrotor UAVs.
• To address the challenge of unknown and uncertain UAV dynamics by employing neural networks (NNs).
• To achieve simultaneous control of all six DOF despite the underactuated nature of the quadrotor.

Main Methods:

• An NN is utilized to learn the complete UAV dynamics online, including aerodynamic friction and blade flapping.
• A novel NN virtual control input scheme is developed to enable control of all six DOF from four inputs.
• An NN observer is introduced to estimate translational and angular velocities, with control laws based on measurable position and attitude.

Main Results:

• Lyapunov theory proves that position, orientation, and velocity tracking errors are semiglobally uniformly ultimately bounded (SGUUB).
• NN weight estimation errors and observer estimation errors are also shown to be SGUUB under bounded disturbances.
• The proposed controller effectively handles unknown nonlinear dynamics and disturbances, demonstrating robust performance.

Conclusions:

• The developed NN-based output feedback controller offers a robust solution for quadrotor UAV control with unknown dynamics.
• The method successfully overcomes the underactuation problem and achieves precise control of all six DOF.
• Simulation results validate the theoretical findings, confirming the effectiveness of the proposed control scheme.