Energy Efficient and Low Latency Computation Offloading via Truly PPO in Mobile Edge Computing Networks with Multi-UAV

Wireless communication technologies such as 5G and 6G are driving the development of intelligent air-ground integrated communications. Unmanned aerial vehicles (UAVs) equipped with edge servers perform well in application scenarios such as disaster relief and large-scale Internet of Things (IoT), substantially alleviating the computational pressure on ground users (GUs). However, most existing task offloading methods are inefficient in dynamic environments due to the dynamically changing communication network structure and limited computational resources. Therefore, we propose a deep reinforcement learning (DRL)-based Fairness Factor Trajectory Selection (FFTS) method to jointly optimize the task offloading decision, resource allocation policy, UAV flight trajectory, and user transmission power to reduce the end-to-end delay and system energy consumption. Specifically, we propose a Fairness Factor Priority (FFP) matching algorithm to optimize task offloading decision and resource allocation strategy. Then, the flight trajectory of UAVs and the transmission power of GUs in each time slot are optimized by a Truly Proximal Policy Optimization (TrulyPPO) algorithm that can limit the trust domain constraints while having a low optimization complexity. The simulation results demonstrate the superiority of our proposed FFTS in terms of delay and energy performance by comparing it with existing benchmark methods based on reinforcement learning.