End-to-End Learning of Deep Visuomotor Policy for Needle Picking