Cooperative Schemes for Joint Latency and Energy Consumption Minimization in UAV-MEC Networks

Area of Science:

• Computer Science
• Artificial Intelligence
• Telecommunications Engineering

Background:

• Internet of Things (IoT) applications demand massive device collaboration, heavy computation, and low latency.
• Unmanned Aerial Vehicle (UAV)-assisted Mobile Edge Computing (MEC) offers flexible, wide-coverage services for User Devices (UDs).
• Optimizing the trade-off between latency and energy consumption in dynamic, large-scale networks is a significant challenge.

Purpose of the Study:

• To jointly optimize association, offloading, and computing resource allocation in UAV-assisted MEC systems.
• To address the inherent trade-off between latency and energy consumption for improved system performance.
• To develop intelligent schemes capable of handling complex and dynamic network environments.

Main Methods:

• Formulated the joint optimization problem as a Partially Observable Markov Decision Process (POMDP).
• Developed two multi-agent deep reinforcement learning (DRL) schemes: Multi-Agent Proximal Policy Optimization (MAPPO) and Closed-Form Enhanced Multi-Armed Bandit (CF-MAB).
• UDs act as independent agents learning from interactions and historical data to maximize individual rewards, fostering implicit collaboration.

Main Results:

• Both proposed DRL schemes demonstrate effectiveness in balancing latency and energy consumption.
• The MAPPO scheme excels in collaborative decision-making for high performance in complex, dynamic environments.
• The CF-MAB scheme provides rapid, independent response decisions by decoupling association from offloading and resource allocation.

Conclusions:

• The proposed multi-agent DRL approaches successfully tackle the challenges of joint optimization in UAV-assisted MEC systems.
• MAPPO and CF-MAB offer distinct advantages for different operational needs, enhancing system performance and efficiency.
• These intelligent schemes provide a robust solution for future large-scale IoT applications requiring efficient resource management.