Text this: Sim-to-Real Deep Reinforcement Learning for Safe End-to-End Planning of Aerial Robots