In recent years, a variety of approaches to robotic control have gained popularity as robots play an increasingly larger role in our everyday lives. In this talk I will give an introduction to modern-day robot learning, covering engineering and research challenges. After establishing the foundations of popular methods, I will present why I believe that end-to-end Machine Learning is the scalable approach for solving robotics problems. Finally, I will talk about some interesting open problems and how data-driven methods can learn efficiently in the real world on a large set of tasks with less supervision.