This study is based on a new approach for an advanced microproduction system or highly flexible production systems where all necessary production and assembly processes are connected in a very flexible way using autonomous mobile transport and handling robots. Each robot has to follow its planned paths while avoiding collisions with other robots. In addition, problem-specific constraints for a defined microproduction system, such as limitations of the velocity and accelerations of the robots, have to be fulfilled. This paper focuses on a two-level model predictive optimizing approach. On a global long-term level, simple dynamic models of the robots are used to compute optimal paths under differential constraints where a safety distance between all robots is achieved. Since many uncertainties and unforeseen events could occur, all robots also use a nonlinear model predictive control approach on a local real-time level. This control approach solves the path following and the collision avoidance problems in parallel, while also taking into account differential constraints of the single robots.