A bi-level mobility-aware deep reinforcement learning approach for fault-tolerant task offloading in vehicular...

Area of Science:

• Intelligent Transportation Systems (ITS)
• Vehicular Edge Cloud Computing (VECC)
• Artificial Intelligence (AI)

Background:

• VECC supports delay-sensitive applications in ITS but faces challenges like high task latency and service failures due to dynamic traffic and network conditions.
• Existing task offloading approaches in VECC often lack robustness against dynamic environmental factors and resource uncertainties.
• Ensuring fault tolerance and minimizing latency are critical for reliable ITS operations.

Purpose of the Study:

• To propose a novel bi-level Deep Q-Network (DQN)-based framework for mobility-aware and fault-tolerant task offloading in VECC.
• To enhance task execution efficiency and reliability by addressing dynamic traffic patterns and uncertain resource availability.
• To reduce task latency and service failures in intelligent transportation systems.

Main Methods:

• A bi-level DQN framework with a level-1 agent for RSU selection and level-2 agents for task allocation and failure recovery (First Result, Recovery Block, Retry).
• Decentralized scheduling by replicating the level-1 DQN across RSUs to eliminate centralized dependency and enhance resilience.
• Simulation using an integrated SimPy/SUMO environment to evaluate performance under heavy and imbalanced traffic conditions.

Main Results:

• The proposed bi-level DQN framework demonstrated significant improvements in total reward, ranging from 7.7% to 37.8%.
• Task failure rates were substantially reduced by 29% to 63% compared to baseline methods (bi-level PPO, Greedy, No-Forwarding).
• The framework proved effective under challenging conditions of heavy and imbalanced traffic.

Conclusions:

• The bi-level DQN approach offers a robust and efficient solution for fault-tolerant task offloading in VECC environments.
• This framework enhances the reliability and performance of intelligent transportation systems by intelligently managing task execution.
• The decentralized, mobility-aware design ensures high accessibility and resilience for critical ITS applications.