AoI Optimal Trajectory Planning for Cooperative UAVs: A Multi-Agent Deep Reinforcement Learning Approach

This paper considers a multiple Unmanned Aerial Vehicles (UAVs)-assisted IoT network, where the UAVs cooperatively collect data packets generated by the IoT devices (IDs) and transmit them to the Base Station (BS) continuously to improve the information freshness, in terms of the age of information (AoI). Particularly, we first formulate multi- UAV distributed cooperative dynamic trajectory planning problem as a decentralized partially observable Markov decision process (Dec-POMDP), where the update arrivals at IDs are stochastic and are unknown to the UAVs. Furthermore, in order to address the challenges arising from unknown environmental dynamics and conflict collision constraints, we devise a multi-agent deep rein-forcement learning (MADRL) based dynamic trajectory planning algorithm. The algorithm leverages the advantages of both the QMIX and Gated Recurrent Unit (GRU) techniques. Finally, simulation results are presented to demonstrate the effectiveness of our proposed algorithm.