This paper addresses the optimization of task offloading and scheduling in vehicular edge computing (VEC) by integrating task prioritization incentives across distributed service areas. To maximize task completion ratios within strict deadlines while minimizing latency and energy consumption, we model the problem as a Partially Observable Markov Decision Process (POMDP). Building on this formulation, we propose a multi-agent deterministic policy gradient approach based on MADDPG, where each agent dynamically selects priority classes, assigns tasks to edge servers, and optimizes CPU frequency allocation using only local observations. Simulation results demonstrate that our MADDPG framework not only meets task prioritization objectives but also achieves performance comparable to, and in some cases exceeding, that of centralized value function-based methods. These findings highlight the ad-vantages of distributed decision-making and localized resource management in complex VEC environments.

This paper primarily focuses on vehicular computation offloading and scheduling based on prioritized incentives. The objective of this work is to maximize task completion within designated deadlines while concurrently minimizing energy consumption and latency. To achieve this, a prioritized Deep Q Network (DQNP) has been proposed to optimize long-term rewards by dynamically selecting a priority class, server, and CPU frequency at each computational step. Simulation results demonstrate that, compared with baseline algorithms, the proposed DQNP can effectively choose tasks, ensuring minimal energy consumption, reduced latency, and improved completion rates, especially for tasks in the highest priority class. Additionally, the findings shed light on how the model alleviates the resource starvation challenge faced by low-priority tasks.