Obstacle avoidance for quadrotors using reinforcement learning and obstacle-airflow interactions