VOILA: Visual-Observation-Only Imitation Learning for Autonomous Navigation