Computation Offloading Based on Multiagent Deep Reinforcement Learning in Vehicle Edge-Cloud Computing

In recent years, the rapid advancement of mobile communication and networking technologies has led to a surge in user demand for increasingly complex applications, imposing substantial computational and communication burdens on the existing networks. While mobile edge computing (MEC) offers task processing capabilities close to users, its limited computational resources remain a significant challenge. To this end, a computation offloading problem within edge-cloud collaborative computing architecture is investigated in this article. Specifically, the problem can be modeled as a partially observable Markov decision process (POMDP). We further propose a computation offloading strategy based on the multiagent deep deterministic policy gradient (MADDPG) and attention mechanism (MADDPG-AT). Extensive simulations demonstrate that MADDPG-AT exhibits strong convergence properties and outperforms several baselines in terms of the tradeoff between energy consumption and latency. Such superiority reduces the total system energy consumption within the maximum tolerable task delay.