Multi-Agent Reinforcement Learning for Cooperative Trajectory Design of UAV-BS Fleets in Terrestrial/Non-Terrestrial Integrated Networks

Aerial base stations (ABSs) have been envisioned as a promising technology toward ubiquitous coverage and seamless high-rate connectivity in sixth-generation (6G) wireless networks. With the inherent mobility but limited communication range, the placement of ABSs should adapt to the time-varying network conditions, e.g., the user distribution and wireless channel state. This letter investigates the cooperative trajectory of UAVs in an integrated terrestrial and non-terrestrial network (TNTN), where unmanned aerial vehicles (UAVs) are deployed as ABSs to supplement the terrestrial macro base station (mBS). We formulate an optimization problem to maximize the number of users with a certain minimum data rate, which is solved using multi-agent reinforcement learning (MARL). Numerical results show that the proposed design is superior to conventional approaches for the cooperative trajectory of UAVs, including K-means clustering-based and single-agent reinforcement learning (SARL)-based methods.