Robust observer backstepping neural network control of flexible-joint manipulator

An output-feedback controller design for flexible-joint manipulators is presented. The proposed controller design consists of a nonlinear Luenberger-type observer, multilayer neural network plant identifier, and controller based on a backstepping framework and variable structure controller. Only link angular positions are measured as outputs. The controller achieves good performance despite the presence of additive external disturbances, unmodeled dynamics, actuator nonlinearities, i.e., deadzone and backlash, and payload changes. Simulation of a two-link flexible-joint manipulator is included.