This work addresses the problem of simultaneously controlling two robotic arms to automatically pick up fruits in an orchard. When considering such a scenario, the design of the controller has to face several challenges: (i) the arms share the same workspace, (ii) the presence of obstacles such as branches and other fruits, and (iii) the system evolves in a dynamic environment, i.e., the positions of the fruits and branches change over time. In this paper, it is proposed to control the two robotic arms relying on a Visual Predictive Control (VPC) scheme. It allows to control the system in a reactive fashion while taking into account a large number of constraints. The global and local models of the VPC scheme as well as constraints dealing with collisions or joint boundaries are presented. Finally, the efficiency of the proposed approach is highlighted with numerous simulation results using the PR2 arms model.