Embodied agents can understand their environment not only through passive observation but can also explore the environment by interaction through their body. We investigate interactive learning approaches that enable intelligent robots to learn models of the functioning of the environment and the effects the robots can have through their body. We research methods for using these models to perform object manipulation or navigation tasks. The models shall be learned from a variety of sensing modalities such as vision, touch and/or proprioceptive sensing, and through interactive exploration of the environment.