Recurrent Neural Network With Fractional Learning-Based Fixed-Time Formation Tracking Constrained Control for a Group of Quadrotors
Hailay Berihu Abebe,Chih-Lyang Hwang,Bor-Sen Chen,Fan Wu,Chau Jan
DOI: https://doi.org/10.1109/access.2021.3083509
IF: 3.9
2021-01-01
IEEE Access
Abstract:In this paper, each agent is modeled by the mechanical motion dynamics with the velocity transformation between quadrotor and world coordinates such that trajectory planning and obstacle avoidance are easily accomplished. It is assumed that at least one follower tracks the leader with a specific position, and the other followers maintain the relative position among each other or the leader. If obstacles hinder the motion of the original formation, a piecewise straight-line formation is employed to avoid these obstacles. To fulfill these tasks under the uncertain dynamics, the recurrent neural network with fractional learning-based fixed-time formation tracking constrained control (RNNFL-FTFTCC) is designed by nonlinear filtering error with dynamic fraction order, time-varying switching gain, and recurrent neural network learning compensation of dynamic lumped uncertainties in each quadrotor. The simulations with the initial formation error, the formation change in a narrow space, and the target point approach validate the effectiveness and robustness of the proposed formation control. Moreover, the comparisons among non-adaptive, RNN, and multilayer perceptron network (MLPN) compensations confirm the effectiveness and efficiency of fractional learning.
computer science, information systems,telecommunications,engineering, electrical & electronic