This paper presents an on-line estimator that incorporates adaptive MIMO radical basis function neural networks (RBFNNs) for model identification of quadrotor unmanned aerial vehicles (UAVs). The inputs and outputs of...
详细信息
ISBN:
(纸本)9781509061914
This paper presents an on-line estimator that incorporates adaptive MIMO radical basis function neural networks (RBFNNs) for model identification of quadrotor unmanned aerial vehicles (UAVs). The inputs and outputs of quadrotor aircrafts can be obtained from dynamic models or real attitude and position sensors. The adaptive learning rate is employed in the gradient descent method for the update of the weights of RBFNNs, and Lyapunov approach guarantees the stability of the global convergence of the modeling errors. The Welsch functions are also employed as the error functions to get rid of the influence from the noise due to disturbances like wind gusts. Simulation results using robotics Toolbox for Matlab verify the effectiveness and robustness of the proposed estimator compared with results of traditional RBFNNs. Experiment results from real aircraft platform show that RBFNNs combining adaptive learning rate and Welsch error functions can approximate the overall system with high accuracy and robustness to disturbances.
暂无评论