[论文解读] Using Empirical Trajectory Data to Design Connected Autonomous Vehicle Controllers for Traffic Stabilization

Yujie Li, Sikai Chen+4 more|arXiv (Cornell University)|Oct 11, 2020

Traffic control and management被引用 13

一句话总结

本文提出了一种面向联网自动驾驶车辆（CAVs）的实时、数据驱动控制器设计，利用NGSIM数据集中的实测轨迹数据，稳定混合交通流。通过将微观交通流模型校准至个体人类驾驶车辆（HDV）行为，并动态调整CAV控制器，该方法显著减少了停车-启动波的传播，提升了混合交通环境下的交通稳定性与安全性。

ABSTRACT

Emerging transportation technologies offer unprecedented opportunities to improve the efficiency of the transportation system from the perspectives of energy consumption, congestion, and emissions. One of these technologies is connected and autonomous vehicles (CAVs). With the prospective duality of operations of CAVs and human driven vehicles in the same roadway space (also referred to as a mixed stream), CAVs are expected to address a variety of traffic problems particularly those that are either caused or exacerbated by the heterogeneous nature of human driving. In efforts to realize such specific benefits of CAVs in mixed-stream traffic, it is essential to understand and simulate the behavior of human drivers in such environments, and microscopic traffic flow (MTF) models can be used to carry out this task. By helping to comprehend the fundamental dynamics of traffic flow, MTF models serve as a powerful approach to assess the impacts of such flow in terms of safety, stability, and efficiency. In this paper, we seek to calibrate MTF models based on empirical trajectory data as basis of not only understanding traffic dynamics such as traffic instabilities, but ultimately using CAVs to mitigate stop-and-go wave propagation. The paper therefore duly considers the heterogeneity and uncertainty associated with human driving behavior in order to calibrate the dynamics of each HDV. Also, the paper designs the CAV controllers based on the microscopic HDV models that are calibrated in real time. The data for the calibration is from the Next Generation SIMulation (NGSIM) trajectory datasets. The results are encouraging, as they indicate the efficacy of the designed controller to significantly improve not only the stability of the mixed traffic stream but also the safety of both CAVs and HDVs in the traffic stream.

研究动机与目标

• 解决异质人类驾驶行为导致的混合交通环境中交通不稳定的问题。
• 开发一种实时、自适应的CAV控制器，以减轻混合交通流中停车-启动波的传播。
• 利用NGSIM数据集中的实测轨迹数据校准个体HDV动力学，实现精确的微观交通建模。
• 将校准后的HDV模型整合进动态CAV控制框架，提升安全性和稳定性。

提出的方法

• 利用NGSIM数据集中的实测轨迹数据校准个体人类驾驶车辆（HDV）动力学，以捕捉异质驾驶行为。
• 开发一种微观交通流（MTF）模型，通过时变参数反映真实世界驾驶行为的变异性，表征个体HDV行为。
• 基于实时更新的校准HDV模型，设计一种联网自动驾驶车辆（CAV）控制器，以适应不断变化的交通状况。
• 实施一种反馈控制策略，利用对HDV动力学的实时估计，抑制交通波增长并稳定流体。
• 采用动态控制器增益调节机制，响应从轨迹数据中提取的交通不稳定度量指标。
• 在仿真中使用NGSIM数据验证控制器，评估其在稳定性与安全性指标方面的性能。

实验结果

研究问题

• RQ1 如何利用实测轨迹数据准确建模混合交通中个体人类驾驶车辆的异质行为？
• RQ2 实时HDV模型校准对CAV控制器在稳定交通流方面性能的影响是什么？
• RQ3 基于数据驱动的CAV控制器在多大程度上可减少混合交通场景中的停车-启动波传播？
• RQ4 基于实时HDV建模的CAV控制器参数动态自适应如何提升交通稳定性和安全性？
• RQ5 与基线策略相比，使用所提控制器在交通稳定性和安全性方面的定量改进程度如何？

主要发现

• 所提控制器在混合交通仿真中显著降低了停车-启动波的振幅与传播速度。
• 利用NGSIM轨迹数据实时校准HDV动力学，使CAV控制策略更加精确且响应更迅速。
• 控制器通过抑制交通波增长，提升交通稳定性，使异质驾驶条件下流体更加平稳。
• 由于减少了急刹车并改善了车距控制，CAV和HDV的安全性指标均得到提升。
• 通过实时适应个体驾驶员动力学，控制器对人类驾驶行为的变异性表现出鲁棒性。
• 定量结果表明，与基线策略相比，行程时间变异性显著降低，混合交通场景下的通行能力提高。