Command filter–based neuroadaptive output feedback control for multi-joint manipulators with prescribed tracking performance

In practical engineering, for nonlinear multi-joint manipulators with unmodeled dynamics, some existing controllers usually require velocity information to achieve an online estimation of model uncertainties; however, it is often difficult to equip with joint angular velocity sensors. To overcome the aforementioned obstacles, this paper proposes a new neural network event-triggered control strategy based on the principle of state observer. Specifically, the neural network is constructed to approximate the unmodeled dynamics in the observer loop, thereby designing a nonlinear observer to recover the joint angular velocities accurately. Then, the nonlinear transformation with a prescribed performance function is exploited to convert the tracking error into an unconstrained form such that the displacement tracking error can be indirectly realized to fall into a prespecified residual set within the user-defined time. Finally, an event-triggered control scheme is designed based on the estimation provided by the neural network in the control loop, which effectively reduces the data transfer between the control unit and the actuating device while guaranteeing the prescribed control performance. More importantly, to further decrease the computational effort of the control algorithm, the command filter with a low-order error compensation system is developed to obtain a structurally simple controller with consideration for the impacts of filtering errors. All signals in the closed-loop system for trajectory tracking are demonstrated to be bounded by applying the Lyapunov function, and the simulation results substantiate the efficacy of the control strategy presented in this study.