Learning to rest: A Q-learning approach to flying base station trajectory design with landing spots