Dependent Task Offloading in Vehicular Edge Computing Using Trajectory-Aware Deep Reinforcement Learning

Vehicular Edge Computing (VEC) enables latency sensitive applications by offloading computation from moving vehicles to roadside units (RSUs). Scheduling dependent tasks in such environments is challenging because vehicles move quickly through RSU coverage zones and connectivity windows are short. We propose a trajectory aware task offloading framework that unifies short-term mobility forecasting with deep reinforcement learning. A mobility-aware transformer predicts the vehicle's future trajectory and derives RSU coverage intervals, while a Graph Attention Network (GAT) based actor-critic, trained via Proximal Policy Optimization (PPO), decides where each task in a directed acyclic graph (DAG) should execute. Experiments with real highway trajectories and a large synthetic DAG library show that the proposed approach consistently outperforms baselines in terms of delay, and energy.