Reinforcement Learning for Versatile, Dynamic, and Robust Bipedal Locomotion Control