Although planning a collision-free path for an object alone has long been solved, planning for safe transportation of objects by robots in presence of obstacles is a challenging task due to high dimensions of the system's composite configuration space. This paper addresses the problem of cooperative object transportation by multiple mobile manipulators through a twolayered Centralized-Decentralized architecture. In the higher (centralized) layer a collision-free global path is planned for the object using a novel sampling-based method called Optimallyconnected Random Tree and without considering the space needed for obstacle avoidance of the robots, and in the lower (decentralized) layer the robots locally avoid obstacles and coordinate their motions with each other in order to reach the object and grasp, transport, and place it at the target position. The proposed method was implemented in Webots™ software for KUKA youBot mobile manipulators and its effectiveness was verified through numerous transportation scenarios.