Worker Classification-Based Task Offloading Scheme for Autonomous Vehicular Networks: Exploiting Deep Reinforcement Learning

Vehicular fog networks play a crucial role in autonomous vehicular networks, which enables the sharing of computing resources of vehicles and thus facilitating the efficient offloading of compute-intensive and latency-sensitive tasks. However, highly dynamic vehicles and stochastic-dependent tasks have posed great challenges for the implementation of vehicular worker-based task offloading. Existing schemes lack comprehensive research into the above issues, resulting in limited adaptability to the dynamic network topology and random task dependencies. To tackle these issues, the task offloading is formulated as a multi-objective optimization problem considering latency, completion rate and cost, and a worker classification based task offloading scheme using deep reinforcement learning is proposed to solve the problem. Specifically, classification specifications of vehicular worker types and input state of workers for DRL model are carefully designed to accommodate dynamic vehicular network topology. Additionally, the cross-job task prioritization scheme considering stochastic task dependencies is proposed to further reduce latency. Simulation results show that the proposed scheme outperforms comparisons in terms of adaptability to dynamic settings and can complete latency-sensitive tasks at a lower cost.