We aim to develop life-long learning machines that can adaptively acquire a wide variety of novel policiesthrough interacting with the real environments.