Scholarlink

<p>Robotaxi services–autonomous, on-demand ride-hailing using self-driving vehicles–are rapidly expanding as technology and shared mobility models advance. However, robotaxi dispatching is challenged by the unpredictable and dynamic nature of urban traffic and fluctuating passenger demand. Existing ride-hailing platforms advocate for “one city, one policy” approaches, which tailor strategies to the unique supply, demand, and influencing factors of each city. They face significant barriers to rapid and large-scale scalability across diverse urban environments. This paper introduces the Cross-city Multi-Agent Policy Transfer (CcMAPT) method to generalize and adapt multi-vehicle dispatching policies for new cities while considering variations in local regulations and demand patterns. By employing matching and distillation techniques, CcMAPT identifies shared environmental dynamics across different cities, thereby complicating rapid scaling and consistent service quality. Experimental results based on real travel data indicate that CcMAPT significantly enhances both the total revenue for robotaxis and the order response rate in Chengdu, Beijing, and Haikou in China. Furthermore, the City Ride-hailing Transportation Simulator (CRTS) has been developed to simulate large-scale, dynamic urban travel environments, thus facilitating the testing and analysis of various vehicle dispatching policies. Validation via the CRTS also demonstrates that CcMAPT possesses robustness across varying city scales, vehicle quantities, and dispatch complexities.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Cross-city robotaxi dispatch in ride-hailing platforms via policy distillation

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月17日

查看全文
  • Lei Tang,
  • Yaling Zhao,
  • Zeyu He,
  • Zhu Wang,
  • Ying Liu,
  • Junchi Ma,
  • Jiahao Yu

摘要

Robotaxi services–autonomous, on-demand ride-hailing using self-driving vehicles–are rapidly expanding as technology and shared mobility models advance. However, robotaxi dispatching is challenged by the unpredictable and dynamic nature of urban traffic and fluctuating passenger demand. Existing ride-hailing platforms advocate for “one city, one policy” approaches, which tailor strategies to the unique supply, demand, and influencing factors of each city. They face significant barriers to rapid and large-scale scalability across diverse urban environments. This paper introduces the Cross-city Multi-Agent Policy Transfer (CcMAPT) method to generalize and adapt multi-vehicle dispatching policies for new cities while considering variations in local regulations and demand patterns. By employing matching and distillation techniques, CcMAPT identifies shared environmental dynamics across different cities, thereby complicating rapid scaling and consistent service quality. Experimental results based on real travel data indicate that CcMAPT significantly enhances both the total revenue for robotaxis and the order response rate in Chengdu, Beijing, and Haikou in China. Furthermore, the City Ride-hailing Transportation Simulator (CRTS) has been developed to simulate large-scale, dynamic urban travel environments, thus facilitating the testing and analysis of various vehicle dispatching policies. Validation via the CRTS also demonstrates that CcMAPT possesses robustness across varying city scales, vehicle quantities, and dispatch complexities.