Posture control of a multi-joint robot based on composition of feedforward joint-torques acquired by iterative learning

This paper aims at generating desired feedforward torque for a specified motion by reuse of time-series torque data acquired by the iterative learning control. The suggested method named motion-scale transformation realizes a motion to an arbitrarily specified posture of a two-DOF planar robot arm on the basis of arithmetical operations of time-series torque of only four motions. The algorism is theoretically presented, and the effectiveness is confirmed in experiments. It is shown by the experimental results that the trajectory tracking errors of angular velocities by the motion-scale transformation tend to be smaller than those by the computed torque method.