A Load-Balancing-Aware Learning Framework for Collaborative UAV-MEC Computation Offloading.

Area of Science:

• Computer Science
• Artificial Intelligence
• Robotics

Background:

• Unmanned Aerial Vehicle (UAV) computing clusters face significant operational constraints due to limited computational power and battery life.
• Simultaneously optimizing low offloading latency, long task endurance, and high cluster efficiency is a complex challenge.

Purpose of the Study:

• To propose a novel framework, MORL-LAPB, for addressing the multi-objective optimization challenges in UAV computing clusters.
• To enhance the operational efficiency and sustainability of UAV clusters through intelligent resource management.

Main Methods:

• Developed a Multi-Objective Reinforcement Learning framework based on Latency and Power Balance (MORL-LAPB).
• Integrated a reward-shaping reinforcement learning algorithm with an evolutionary mechanism for closed-loop optimization.
• Implemented a joint optimization strategy balancing task allocation and energy depletion rates across UAV nodes.

Main Results:

• MORL-LAPB significantly reduces offloading latency compared to baseline methods (RSO, NSO, DRLSO).
• The framework effectively extends the duration of task execution and improves overall cluster energy efficiency.
• Demonstrated efficient identification of Pareto optimal trade-offs for dynamic resource scheduling.

Conclusions:

• MORL-LAPB offers a robust solution for optimizing UAV computing clusters under strict multi-objective constraints.
• The framework provides flexible adaptability and promotes long-term sustainability in diverse operational scenarios.
• Achieved a balance between service delays, energy depletion rates, and cluster efficiency for enhanced UAV operations.