Scholarlink

<p>To address the current limitations of autonomous driving systems in naturally imitating human driving behavior, this study proposes a human-like behavior decision-making model that incorporates driver expectations. Driving expectations refers to a driver's subjective preferences regarding desired speed and following distance. By quantifying the deviation between these expectations and the actual driving state, cumulative dissatisfaction values for speed and car-following distance were introduced as extended features of the model. Typical behavioral and environmental features were selected from the HighD real-world highway trajectory dataset, and a combination of Spearman correlation and random forest importance analysis was employed for feature selection. On this basis, a BiLSTM-attention decision-making model was developed by integrating a bidirectional long short-term memory network and an attention mechanism. In addition, a minimum safety spacing model was incorporated to impose safety constraints on lane-changing behavior. The model was trained and evaluated using the HighD dataset. Experimental results show that it achieved an average prediction accuracy of 94.22% for driving behaviors in highway scenarios, outperforming other models while using fewer features.</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

Human-Like Behavior Decision-Making for Autonomous Driving Based on Driving Expectations

  • 期刊论文

  • 发表日期：

    2025年9月22日

查看全文
  • Shuen Zhao,
  • Deyang Zhu,
  • Maojie Zhao

摘要

To address the current limitations of autonomous driving systems in naturally imitating human driving behavior, this study proposes a human-like behavior decision-making model that incorporates driver expectations. Driving expectations refers to a driver's subjective preferences regarding desired speed and following distance. By quantifying the deviation between these expectations and the actual driving state, cumulative dissatisfaction values for speed and car-following distance were introduced as extended features of the model. Typical behavioral and environmental features were selected from the HighD real-world highway trajectory dataset, and a combination of Spearman correlation and random forest importance analysis was employed for feature selection. On this basis, a BiLSTM-attention decision-making model was developed by integrating a bidirectional long short-term memory network and an attention mechanism. In addition, a minimum safety spacing model was incorporated to impose safety constraints on lane-changing behavior. The model was trained and evaluated using the HighD dataset. Experimental results show that it achieved an average prediction accuracy of 94.22% for driving behaviors in highway scenarios, outperforming other models while using fewer features.