A simple alternative to neural network control scheme for robot manipulators

Recent research results have shown that neural network techniques are eeective in compensating highly nonlinear uncertainties in the robot model where computed torque method is used for robot motion control. One excellent work was reported by Ishiguro et:al 1]. The propose of this note is to present a simple alternate solution to the same control problem which eliminates the need of a neural network. The solution is based on the disturbance rejection technique by the authors 2]. Computer simulations show that the alternate control method works better. It is a well established fact that model-based computed-torque control scheme is the basic technique for robot control system design. But it is also a common knowledge that the scheme has poor robustness when uncertainties exist in the robot model. To improve the robustness, application of both adaptive control techniques and neural network control techniques have been proposed. The latter approach has been actively investigated in recent years. Eliminating the eeects of model uncertainties by adaptive control scheme is to estimate the robot parameters on-line. On the other hand, a disturbance rejection torque is generated on-line by a neural network to cancel out the eeects of uncertainties in robot dynamics 1]. Both of these approaches require extensive computation. The purpose of this note is to examine the basic idea behind the neural network scheme proposed by Ishiguro et:al 1], and to present an extremely simple alternate scheme to solve the same problem. The robot control scheme proposed in 1] can be summarized as follows: Given the robot manipulator dynamics as