Comparative Study of LQR and Neural Network Controllers for Qadcopter Roll Angle Stabilization

Abstract

This paper presents a comparative analysis between a conventional Linear Quadratic Regulator (LQR) and a neural network-based approach for quadcopter roll angle stabilization. While the LQR controller delivers optimal performance in nominal conditions, its accuracy deteriorates when subjected to model uncertainties and persistent external disturbances, causing steady-state deviations.

To mitigate these issues, we develop a neural network controller trained via imitation learning. The proposed architecture inherently learns an integral action, enabling perfect rejection of constant disturbances. Extensive simulations show that our neural controller achieves superior performance in realistic operating conditions while maintaining similar stability margins to LQR in ideal cases.

This study demonstrates how machine learning techniques can enhance classical control paradigms for aerial robotics. The results suggest that neural network controllers offer a viable path toward more adaptive and robust autonomous drone systems in challenging environments.