A Closed-Loop Multi-perspective Visual Servoing Approach with Reinforcement Learning