Scholarlink

Human-like driver modeling has recently emerged as an important research focus in autonomous vehicles and intelligent transportation systems. We present a unified driver model that captures two types of driving maneuvers present in every driver: a commonality model for general driving skills, and a personality model for individual driving characteristics. The commonality model utilizes a networked PID controller. Using the distal learning control method, the commonality model can adapt itself to specific scenarios, while the personality model is initially developed from real-world driving behavior data and further refined alongside the commonality model. The FTP-72 driving cycle serves as the reference speed profile. Simulation results from driving cycle tests demonstrate continuous improvement in speed control performance through self-learning, improving the precision of speed control from [−5, + 5] mph to [−3, + 3] mph, while retaining the original driver's unique driving style.

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

hDriver: A Human-Like Driver Model Capable of Style-Retaining and Self-Learning for Better Speed Control

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Xiangguang Ma,
  • Xuanzhong Chen,
  • Jie Hu

摘要

Human-like driver modeling has recently emerged as an important research focus in autonomous vehicles and intelligent transportation systems. We present a unified driver model that captures two types of driving maneuvers present in every driver: a commonality model for general driving skills, and a personality model for individual driving characteristics. The commonality model utilizes a networked PID controller. Using the distal learning control method, the commonality model can adapt itself to specific scenarios, while the personality model is initially developed from real-world driving behavior data and further refined alongside the commonality model. The FTP-72 driving cycle serves as the reference speed profile. Simulation results from driving cycle tests demonstrate continuous improvement in speed control performance through self-learning, improving the precision of speed control from [−5, + 5] mph to [−3, + 3] mph, while retaining the original driver's unique driving style.