Multi-Modal Mutual Information (MuMMI) Training for Robust Self-Supervised Deep Reinforcement Learning | IEEE Conference Publication | IEEE Xplore