Reconstructing Telesurgical Manipulator Pose Via Reinforcement Learning