Learning of inverse-kinematics for robot using high dimensional neural networks

In this paper, we describe a position control for robots using high dimensional neural networks. The simultaneous perturbation optimization method is used for the learning rule of the high dimensional neural networks. Then the high dimensional neural networks learn inverse kinematics of the robots. Control objectives are two dimensional SCARA robot and three dimensional robot. Tip of these robots are controlled by a complex-valued neural network and a quaternion neural network, respectively. Some simulation results are shown to confirm feasibility of these high dimensional neural networks as robot controller.