Due to the limited bandwidth of Roadside Units (RSUs) deployed in drive-thru networks, vehicles entering the network coverage with data requests have to contend for the access to the data service provided by RSUs. In order to maximize the vehicle utility, efficient access schemes are indispensable at the vehicles’ side. This paper studies the optimal access control of vehicles in multi-agent drive-thru systems. In such networks, each vehicle, acting as an independent agent, can take an access decision that could potentially maximize the individual utility based on its own observations of the instantaneous environment states. Consequently, the decision of one vehicle will influence those of others, making environment states only partially observable at the vehicles’ side and complicating the optimal access design. To tackle this coupling decision issue, we first formulate the optimization problem as a finite Markov Decision Process (MDP). Then, we propose a distributed access algorithm that combines the statistic learning method and the dynamic programming technique. With the proposed algorithm, missing vehicle states and related transition probabilities will be estimated by vehicles. The optimization problem is recursively solved by applying the dynamic programming technique. Simulation results are provided to show the significant improvement achieved by the proposed algorithm on multiple performance metrics. The convergence of the algorithm is numerically confirmed, verifying the stability of our approach.

中文翻译：

---

多代理V2I系统中车辆的分布式通道接入方案

由于部署在得来速网络中的路侧单元 (RSU) 的带宽有限，带着数据请求进入网络覆盖范围的车辆必须竞争对 RSU 提供的数据服务的访问。为了最大限度地发挥车辆的效用，有效的访问方案在车辆一侧是必不可少的。本文研究了多智能体免下车系统中车辆的最优访问控制。在这样的网络中，每辆车作为一个独立的代理，可以根据自己对瞬时环境状态的观察来做出访问决策，该决策可能使个人效用最大化。因此，一辆车的决定会影响其他车辆的决定，使得环境状态只能在车辆一侧观察到，并使最佳访问设计复杂化。为了解决这个耦合决策问题，我们首先将优化问题表述为有限马尔可夫决策过程（MDP）。然后，我们提出了一种将统计学习方法和动态规划技术相结合的分布式访问算法。使用所提出的算法，车辆将估计丢失的车辆状态和相关的转移概率。优化问题是通过应用动态规划技术递归解决的。仿真结果显示了所提出的算法在多个性能指标上取得的显着改进。算法的收敛性得到了数值验证，验证了我们方法的稳定性。我们提出了一种结合了统计学习方法和动态规划技术的分布式访问算法。使用所提出的算法，车辆将估计丢失的车辆状态和相关的转移概率。优化问题是通过应用动态规划技术递归解决的。仿真结果显示了所提出的算法在多个性能指标上取得的显着改进。算法的收敛性得到了数值验证，验证了我们方法的稳定性。我们提出了一种结合了统计学习方法和动态规划技术的分布式访问算法。使用所提出的算法，车辆将估计丢失的车辆状态和相关的转移概率。优化问题是通过应用动态规划技术递归解决的。仿真结果显示了所提出的算法在多个性能指标上取得的显着改进。算法的收敛性得到了数值验证，验证了我们方法的稳定性。仿真结果显示了所提出的算法在多个性能指标上取得的显着改进。算法的收敛性得到了数值验证，验证了我们方法的稳定性。仿真结果显示了所提出的算法在多个性能指标上取得的显着改进。算法的收敛性得到了数值验证，验证了我们方法的稳定性。