Learning and control model of arm posture

We will introduce a learning and control model of arm posture which can be thought useful as a control model for a humanoid-robot, such as a muscle-driven robot in terms of generating proper tensions to each muscle. It is difficult to generate proper tensions for controlling arm Posture because the problem is redundant, which means that the number of ways of choosing tensions is a lot even in generating the same joint torque. In simulation of using the model, we got proper and energy efficient tensions needed to maintain the arm in one's goal positions. The model used the Actor-Critic method and the feedback-error-learning scheme, which are considered as a control model in the brain that humans use.