Real-time Trajectory Tracking of a Quadrotor using Adaptive Backstepping Controller and RNN based Uncertainty Observer

Abstract

This paper presents an approach for position and attitude control of a quadrotor using adaptive backstepping technique along with an uncertainty observer via Recurrent Neural Network (RNN). The quadrotor dynamics are expressed as two subsystems, namely translational and rotational, on which the backstepping control law has been developed. In comparison with feedforward neural networks, RNN has better dynamic characteristics and approximation capabilities. Therefore, an RNN based uncertainty observer has been employed to accommodate the system uncertainties as well as the unknown external disturbances. The proposed controller consists of two parts - an adaptive backstepping based controller that contains an RNN observer and a robust controller to deal with the approximation error induced by the RNN. The RNN parameters have been updated via an update law based on Lyapunov stability theory in an online manner where the overall system stability is also guaranteed. The proposed approach has been implemented in simulations for trajectory tracking of the quadrotor in the presence of parametric uncertainties and external disturbances. Also, the hardware results are presented to show the effectiveness of the proposed approach on DJI Matrice 100.