Utilizing the high mobility of UAVs to collaborate with wireless communication networks for data collection is considered an effective technology. This chapter investigates a multi-UAV-assisted data collection system with no-fly zones (NFZs), where UAVs take off from an initial location to collect data from mobile ground users (GUs) while avoiding entering NFZs and preventing collisions. The objective of this work is to minimize the whole data collection time via optimizing the 3D trajectories of UAVs and the scheduling of mobile GUs. On the one hand, the presence of NFZs makes the formulated problem non-convex. In addition, the mobility of users results in a dynamically changing system environment, making it difficult to solve using traditional methods. To tackle the above challenges, we begin by converting the problem into a Markov Decision Process (MDP) and introduce a multi-agent federated reinforcement learning (MAFRL) approach to optimize the long-term goal. Specifically, we implement a multi-step propagation technique alongside a dueling network architecture to enhance the neural network used for agent training, aiming to speed up convergence and increase overall stability. Finally, simulations validate the effectiveness of the method in practical scenarios.