Methods and apparatus, including computer program products, are described for managing tube orientation in a food packaging machine, wherein the food packaging machine comprises a plurality of sub-systems. One or more variable values are received, which indicate measurements by the food packaging machine of one or more physical parameters in one or more of the sub-systems, the one or more physical parameters affecting tube orientation. One or more control parameter values are determined for one or more of the sub-systems, by processing the received variable values using a reinforcement learning model and a local control model. One or more control parameters of the one or more sub-systems are adjusted in accordance with the determined control parameter values.