I'm a beginner in this topic and are learning how to build an artificial neural network and different types of training associated with them.
Right now, I'm trying to figure out self learning. For example, there is a robot arm consisting of two joints and three parts. The objective is the distance from target (I need the arm to touch the target), neural network inputs are the coordinates of the target, the output is the joint angles of the arm. But, the error is the distance of hand to target.
Since I'm using an error-back-propagation feed-forward neural network, its error is only the output and desired output, but I need it to find the desired output (target joint angle) itself and take the error from distance, not joint angle. How can this be done?
More clearly, what kind of feed-back scheme is there for a back-propagation feed-forward ANN for self-learning of touching a coordinate with robot arm?