We present a framework for visual action planning of complex manipulation\ntasks with high-dimensional state spaces such as manipulation of deformable\nobjects. Planning is performed in a low-dimensional latent state space that\nembeds images. We define and implement a Latent Space Roadmap (LSR) which is a\ngraph-based structure that globally captures the latent system dynamics. Our\nframework consists of two main components: a Visual Foresight Module (VFM) that\ngenerates a visual plan as a sequence of images, and an Action Proposal Network\n(APN) that predicts the actions between them. We show the effectiveness of the\nmethod on a simulated box stacking task as well as a T-shirt folding task\nperformed with a real robot.\n