.png){kind=icon}
Return
Adaptive multi-objective cooperative optimization algorithm for ground-air collaborative mobile edge intelligence offloading
.png){kind=icon}
PRE
AI
DOI：10.1016/j.vehcom.2026.101044.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The efficacy of single unmanned aerial vehicles (UAVs) in complex task scenarios is largely limited by their constrained computational resources and battery endurance. Ground-air collaborative systems have thus emerged as vital solutions to mitigate these limitations. However, existing deep reinforcement learning (DRL)-based computational offloading approaches for ground-air cooperation primarily focus on single-objective optimization, making it difficult to simultaneously accommodate dynamic task requirements and varying resource states, and lacking effective multi-objective collaborative optimization capabilities. To address this issue, this work proposes a Hierarchical Multi-Objective Deep Reinforcement Learning (HMODRL) algorithm for the joint optimization of computational offloading and task scheduling. HMODRL adopts a two-layer framework: the lower layer employs a Transformer-Enhanced Multi-Objective Twin Delayed Deep Deterministic Policy Gradient (TMO-TD3) algorithm, which dynamically adjusts objective weights to construct an approximate Pareto front, thereby providing an intuitive representation of trade-offs among multiple objectives; the upper layer leverages a Deep Contextual Bandit (DCB) algorithm to perceive real-time environmental states and adaptively predict and adjust the objective weights of the lower layer, enabling dynamic optimization for heterogeneous UAVs performing diverse tasks. Simulation results demonstrate that HMODRL significantly outperforms benchmark algorithms in overall optimization performance. Moreover, the proposed adaptive weighting mechanism notably enhances system adaptability to dynamic environmental changes and diverse task requirements, improving the UAV task completion rate by 23.6%.
Keywords:
Computation offloading
Task scheduling
Ground-air collaborative
Deep reinforcement learning
Multi-objective optimization
Journal
IF:
6.5
Papers:
819
Citations:
3.2K
Organization
View more organizations
Cited Papers
No cited papers available