Joint Cooperative Computation Offloading and Trajectory Optimization in Heterogeneous UAV-Swarm-Enabled Aerial Edge Computing Networks

Aerial edge computing (AEC) networks, which employ multiple unmanned aerial vehicles (UAVs) as mobile-edge computing servers, have emerged as a promising solution to provide computation offloading services, especially in scenarios where the coverage of existing infrastructures is limited for wireless networks. Recently, there has been a growing focus on leveraging UAVs with diverse computing capabilities to enhance the performance of AEC networks through cooperative computing. However, heterogeneity and cooperation introduce a higher degree of coupling between trajectory planning and computing offloading strategy for AEC networks. In particular, the joint decision making in an AEC network needs to balance minimizing the distance between UAVs and access users and enabling collaborative offloading. And this must be done while considering the time-varying computation requirements and the long-term impact on system performance. To address the aforementioned challenges, we formulate an optimization problem to design a joint dynamic cooperative computation offloading and trajectory optimization scheme for the AEC network. The complexity arises from the problem’s nature as a mixed-integer nonlinear program. To tackle this challenge, we propose a multiagent deep reinforcement learning algorithm based on QMIX. We leverage both theoretical analysis and an action branching architecture to reduce the complexity of our proposed deep reinforcement algorithm. Simulation results demonstrate a substantial performance improvement over the benchmarks, affirming the effectiveness of our complexity reduction approach.