Robot learning uses advanced machine learning and statistical methods to improve the effectiveness of algorithms and models used in robotics, including low-level control, dynamics models and high-level planning. Having such autonomous learning for physical systems has been a long-standing vision of robotics, artificial intelligence, and cognitive sciences. In the past two decades, the Robot Learning community has developed a large repertoire of learning techniques that enable robots to acquire new skills from data. This book chapter aims to provide a straightforward and intuitive introduction to the approaches that enabled these methods, starting from optimal control theory, and incorporating ideas from machine learning that facilitate model-based and model-free reinforcement learning and imitation learning. To enable practical learning on real robotic platforms, we describe inductive biases for robotic learning, such as movement primitives; and how they are used for sample-efficient, data-driven control.