Exponential-Type Variable Parameter Finite-Time ZNN for Solving Joint-Drift-Free Inverse Kinematics of Redundant Robots

This paper proposes an exponential-type variable parameter finite-time zeroing neural network (EVPFTZNN) that incorporates a finite-time bounded activation function (BAF). The proposed variable-parameter function exhibits exponential growth dependent on the error signal within prescribed bounds, while its boundary parameters remain fully configurable to accommodate specific application requirements. For example, most parameters are time-varying and are subject to physical constraints, meaning they cannot exceed certain maximum values. Consequently, unbounded activation functions and variable parameters that grow infinitely with time may be somewhat challenging in actual hardware implementation. The EVPFTZNN model studied in this paper is capable of circumventing this problem. Theoretical derivation proves the finite-time convergence of the EVPFTZNN and provides the analysis of its disturbance rejection capability. Furthermore, to verify the practicality of the EVPFTZNN, this paper applies it to solve the joint-drift-free inverse kinematics problem of redundant robots. Compared with several existing ZNNs, computer simulations and physical experiments validate the superior performance of the EVPFTZNN for solving joint-drift-free inverse kinematics of redundant robots.