Scholarlink

<p>Vehicle trajectory prediction is a crucial foundation for behavioral decision-making and path planning in autonomous driving systems. To address the underutilization of temporal information for the target vehicle and the complexity of spatial interactions between vehicles in current trajectory-prediction models, this paper proposes a trajectory-prediction model called Driving Habits–Multi-head Graph Attention (DH-MHGA). The model consists of a Driving Habits Extraction Module (DHEM), a Driving Habits-Temporal Fusion Encoder (DHTFE), and a Multi-head Graph Attention Spatial Decoder (MGASD). Specifically, the DHEM extracts driving habits from existing vehicle trajectories. Then, the DHTFE combines the observed historical trajectory with the driving habits and encodes the resulting data into high-dimensional features. Finally, the MGASD generates multimodal prediction trajectories by extracting spatial interaction features between the target vehicle and adjacent vehicles and selecting the driving habits with the highest confidence. This study evaluates DH-MHGA using freeway scenarios from the Next Generation Simulation (NGSIM) dataset. Compared with CS-LSTM, DH-MHGA reduces the minADE by 35.6% and the minFDE by 45.9%.</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 vehicle trajectory prediction based on driving habits and multi-head attention mechanism

  • 期刊论文

  • 发表日期：

    2026年7月27日

查看全文
  • Lujiao Li,
  • Yongbin Hu,
  • Zifeng Liu,
  • Jiafu Wang,
  • Long Chen

摘要

Vehicle trajectory prediction is a crucial foundation for behavioral decision-making and path planning in autonomous driving systems. To address the underutilization of temporal information for the target vehicle and the complexity of spatial interactions between vehicles in current trajectory-prediction models, this paper proposes a trajectory-prediction model called Driving Habits–Multi-head Graph Attention (DH-MHGA). The model consists of a Driving Habits Extraction Module (DHEM), a Driving Habits-Temporal Fusion Encoder (DHTFE), and a Multi-head Graph Attention Spatial Decoder (MGASD). Specifically, the DHEM extracts driving habits from existing vehicle trajectories. Then, the DHTFE combines the observed historical trajectory with the driving habits and encodes the resulting data into high-dimensional features. Finally, the MGASD generates multimodal prediction trajectories by extracting spatial interaction features between the target vehicle and adjacent vehicles and selecting the driving habits with the highest confidence. This study evaluates DH-MHGA using freeway scenarios from the Next Generation Simulation (NGSIM) dataset. Compared with CS-LSTM, DH-MHGA reduces the minADE by 35.6% and the minFDE by 45.9%.