Simoun: Synergizing Interactive Motion-appearance Understanding for Vision-based Reinforcement Learning | IEEE Conference Publication | IEEE Xplore