Sensorimotor coordination in humanoid robots is the key to accomplish realistic human behavior. Paradigmatic tasks in sensorimotor coordination include force and impedance control, whole-body coordination during physical interaction with the environment, point-to-point reaching movements, and grasping visually identified objects. To tackle these problems a variety of sensors and methods have to be integrated, including vision, force and touch, hand-coded models, and machine learning. This requires a careful balance of the a priori design effort in order to properly manage the complexity of data collection for learning and the overall performance of the robotic system. In this chapter we treat the implementation of these techniques on the well-known iCub humanoid robot.