MULTILAYER NEUROADAPTIVE CONTROL OF TWO DOF ROBOT WITH DEADZONE

Abstract

Robots generally su er from nonlinearities such as deadzone, backlash and deadband. The inherent coupling e ect between links of the robot along with nonlinearities puts consid erable challenge to control the position of end e ector. A two link manipulator is a basic robot in which most of the controllers are implemented to get a fair idea of the control performance. In this paper we use a adaptive neural network for controlling a two link rigid planar manipulator with deadzone at input. The deadzone at the input is not static but dy namic. To remove the e ect of deadzone, inverse deadzone is used which just adds an o set to the control input. The o set is greater than or equal to the deadzone width and hence there could be some extra input going to the actuator which is modeled as disturbance in the system. Performance of the Multilayer Neuroadaptive (MN) controller is compared with typical Single Layer Neuroadaptive controller, PID and classical Adaptive Controller in sim ulation. Error norms and other parameters are calculated for di erent control performance are given to compare quantitatively.