We cannot make a redundant manipulator determine the unique posture with only position information. But it is possible if we use the performance function of the posture. However the solution is difficult in analysis because its equation contains the complex non-linear functions. There are alternative solutions used numerical computations, however these ways are needed many seconds and converge out of mobile area with manipulator on default value. In this paper we present that the supervising signals are made with numerical computations and those signals are used to learn a neural network. And we prove this method useful because the designed neural network has given the optimum posture according to only position information in the simulation.
