Scholarlink

<p>Accurate trajectory prediction is essential for autonomous vehicle decision-making, especially in urban settings where motorcycles significantly impact. Traditional models often ignore the flexibility and unpredictability of motorcycles. To address this, we introduce a new trajectory prediction model that incorporates motorcycle interactions. This model utilizes Gated Recurrent Units to capture motorcycle acceleration and employs graph structures to encode environmental data, focusing on the lateral distance between motorcycles and autonomous vehicles to develop comprehensive spatiotemporal features. We apply a cross-attention mechanism to create a driving intention probability model, producing a heatmap of potential vehicle positions in the near future. A multi-layer perceptron acts as the decoder to deliver multimodal trajectory predictions. model outperforms existing ones, achieving improvements of 36.89%, 38.49%, and 37.95% in min ADE, min FDE, and MR, respectively, highlighting its effectiveness in this context.</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

Multimodal trajectory prediction model of intelligent vehicle under the influence of motorcycle interaction

  • 期刊论文

  • 发表日期：

    2025年10月13日

查看全文
  • Zhiyu Yang,
  • Xue Yang,
  • Lingcong Yuan,
  • Xiaolong Tan

摘要

Accurate trajectory prediction is essential for autonomous vehicle decision-making, especially in urban settings where motorcycles significantly impact. Traditional models often ignore the flexibility and unpredictability of motorcycles. To address this, we introduce a new trajectory prediction model that incorporates motorcycle interactions. This model utilizes Gated Recurrent Units to capture motorcycle acceleration and employs graph structures to encode environmental data, focusing on the lateral distance between motorcycles and autonomous vehicles to develop comprehensive spatiotemporal features. We apply a cross-attention mechanism to create a driving intention probability model, producing a heatmap of potential vehicle positions in the near future. A multi-layer perceptron acts as the decoder to deliver multimodal trajectory predictions. model outperforms existing ones, achieving improvements of 36.89%, 38.49%, and 37.95% in min ADE, min FDE, and MR, respectively, highlighting its effectiveness in this context.