Reinforcement Learning With Vision-Proprioception Model for Robot Planar Pushing