Visible to the public Learning Adaptive Representations for Robust Mobile Robot Navigation from Multi-Modal Interactions