Obstacle-Aware Quadrupedal Locomotion With Resilient Multi-Modal Reinforcement Learning