Adaptive neural finite-time tracking control based on modified command-filtered backstepping method for MIMO nonlinear systems

Abstract

This paper develops an adaptive neural finite-time tracking control scheme based on modified command-filtered backstepping control method (MCFBC) for multi-input and multi-output (MIMO) nonlinear systems. Firstly, by introducing some constant matrices, command-filtered backstepping control method (CFBC) is modified. Compared with CFBC, MCFBC ensures that virtual control signals are linearized, which lowers the complexity of the controller such that the control performance is elevated. Secondly, different from CFBC, MCFBC does not directly use the spectral boundedness of control directions in stability proof any more. Thirdly, lumped uncertainties are approximated by neural network (NN) and a more generalized inequality is proposed to surmount the technical difficulties of finite-time stability analysis. Fourthly, the singularity problem is circumvented. Command filters are introduced to remove the repeated differentiation of pseudocontrol signals. An error compensation system is constructed to lower the adverse effect from filter errors. This proposed controller guarantees that all signals in the closed-loop system converge to a bounded region within finite time and the system output can follow the given signal with a considerably small tracking error. Finally, flexible joint manipulations are used to validate this control scheme.