Two recurrent neural networks for local joint torque optimizationof kinematically redundant manipulators

摘要

This paper presents two neural network approaches to real-timenjoint torque optimization for kinematically redundant manipulators. Twonrecurrent neural networks are proposed for determining the minimumndriving joint torques of redundant manipulators for the eases withoutnand with taking the joint torque limits into consideration,nrespectively. The first neural network is called the Lagrangian networknand the second one is called the primal-dual network. In bothnneural-network-based computation schemes, while the desirednaccelerations of the end-effector for a specific task are given to thenneural networks as their inputs, the signals of the minimum drivingnjoint torques are generated as their outputs to drive the manipulatornarm. Both proposed recurrent neural networks are shown to be capable ofngenerating minimum stable driving joint torques. In addition, thendriving joint torques computed by the primal-dual network are shownnnever exceeding the joint torque limits