Recently, Unmanned Aerial Vehicles (UAVs) have attracted a great deal of attention in academic, civilian and military communities as prospective solutions to a wide variety of applications. The use of cooperative UAVs has received growing interest in the last decade and this provides an opportunity for new operational paradigms. In this paper, the problem of formation reconfiguration for a group of N cooperative UAVs in an obstacle-loaded environment is solved using decentralized Learning Based Model Predictive Control (LBMPC). The formation of the multiple cooperative UAVs respects the general rules of flocking known as Reynold's rules. Each UAV is required to avoid collision with nearby flockmates, attempts to match the velocity of other team members and attempts to stay close to other team members respecting the desired formation. When static obstacles appear, the UAVs are required to steer around the obstacle or pass through avoiding collision with the obstacles or with each other. A state transformation algorithm is applied to linearize the UAV dynamics generating a linear system allowing the implementation in real life. Our main contribution in this paper lays in solving the formation reconfiguration problem for a group of cooperative UAVs forming a desired formation using decentralized LBMPC in the presence of obstacles.