High precision position control by Cartesian trajectory feedback and connectionist inverse dynamics feedforward

Abstract

An optimal Cartesian trajectory determination coupled with a connectionist approach to perform the dynamics inversion is presented. This method uses a recurrent calculation of the optimal Cartesian trajectory function in order to drive the arm to the desired position and velocity in the desired time. Using this principle of dynamic optimality it is shown that it is possible to achieve the goal with an arbitrary precision even though the inverse dynamics transformation is only an approximation obtained by a neural network. The analysis of simulated control strategy shows that the relative position error for a start-stop movement follows a high inverse power law with respect to the number of feedback control steps. This result indicates that it is practical to control a manipulator to an arbitrary degree of precision by using a neural network whose transformation has a relatively low precision.