The MOZART project aims to advance the state-of-the art in robotic handling and manipulation of soft and heterogenous objects. The approach we take is a radical departure from existing manipulator approaches as we will develop a new concept for manipulation through the use of deformable manipulation surfaces supported by AI-powered control and learning tools. The manipulation surface is able to change its curvature locally and through this manipulates objects on the surface. The manipulation surface has three constituent elements. A conventional modular electronics layer responsible for communication, power, and computation, an actuation layer based on origami mechanisms, and finally, a soft poro-elastic sensing skin. The hardware development is complemented by a distributed control application programming interface which is built on cellular automata and deep learning. The framework includes basic and coordination primitives that allow for manipulation of individual objects and groups of objects, respectively. For demonstration we have chosen three target demonstrators in the food handling industry which are descaling fish, sorting of chicken and presentation of chicken. The final pillar of the project is an integrative-interdisciplinary social sciences and humanities approach that tackles the transition of the role of the operators today to more high-skilled jobs in the future. It also addresses the general ethical challenges surrounding food production to ensure sustainable technology development in terms of social and ethical as well as ecological consequences. The core research is complemented by a comprehensive dissemination and communication package that targets all relevant stakeholders. Finally, a clear exploitation plan including standardisation is described which will transition the AUTOMATs from TRL 5 at the end of the project to a product ready to impact the food packing industry four year after the end of the project.