Abstract
Task offloading to edge servers enables resource-constrained vehicles to run advanced AI services. However, limited edge capacity often leads to overload, especially when traffic surges in specific regions such as during rush hours or major events. Existing vehicular offloading schemes confine cooperation within fixed or neighbor-based scopes, causing severe performance degradation when all servers in a region become saturated. Expanding the cooperation range can alleviate overload but introduces a new cost - system-wide energy consumption. An effective framework must therefore adapt its cooperation scope to real-time network conditions while jointly optimizing the success-energy trade-off. This paper proposes LARA-PPO, a two-layer offloading framework that fulfills both requirements. We first formulate a joint optimization problem to maximize task success while minimizing total energy. The first layer, LARA (Lightweight Adaptive Resource-aware Aggregation), dynamically forms cooperation groups by expanding or shrinking their scope according to real-time resource conditions through decentralized, range-limited message exchanges. Within each group, a Proximal Policy Optimization (PPO)-based reinforcement learning agent jointly determines the offloading fraction and task placement to balance success and energy. Simulation results show that LARA-PPO surpasses the success-only scheme with higher task success rates and about 20% lower total energy consumption. Under localized overload, it further outperforms static-region and neighbor-based cooperation by 15% and 25%, respectively, demonstrating robust adaptability through dynamic group expansion.
| Original language | English |
|---|---|
| Title of host publication | ICC 2026 - IEEE International Conference on Communications, Proceedings |
| Publisher | Institute of Electrical and Electronics Engineers Inc. |
| ISBN (Electronic) | 9798319542090 |
| DOIs | |
| State | Published - 2026 |
| Event | 2026 IEEE International Conference on Communications, ICC 2026 - Glasgow, United Kingdom Duration: 24 May 2026 → 28 May 2026 |
Publication series
| Name | IEEE International Conference on Communications |
|---|---|
| ISSN (Print) | 1550-3607 |
Conference
| Conference | 2026 IEEE International Conference on Communications, ICC 2026 |
|---|---|
| Country/Territory | United Kingdom |
| City | Glasgow |
| Period | 24/05/26 → 28/05/26 |
Bibliographical note
Publisher Copyright:© 2026 IEEE.
UN SDGs
This output contributes to the following UN Sustainable Development Goals (SDGs)
-
SDG 7 Affordable and Clean Energy
Keywords
- Cooperative computing
- edge computing
- reinforcement learning
- task offloading
Fingerprint
Dive into the research topics of 'LARA-PPO: A Joint Optimization Framework for Success-Energy Trade-off in Vehicular Edge Task Offloading'. Together they form a unique fingerprint.Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver