Scholarlink

This study investigated the impact of various state representation methods on vehicle decision-making at unsignalized intersections using deep reinforcement learning. We proposed a novel hybrid state representation framework that combines an attention mechanism with collision time prediction, substantially enhancing autonomous decision-making. Unlike traditional techniques, our approach fused complementary features to improve obstacle avoidance and incorporates an action masking module to optimize traffic efficiency. To demonstrate the practicality and efficiency of the proposed method, extensive simulation experiments were conducted, and the results validate its effectiveness.

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

Hybrid State Representation with Attention and Collision Time Prediction for Autonomous Vehicle Decision-Making at Unsignalized Intersections

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Ruming He,
  • Junran Xie,
  • Penghui Shang,
  • Yifan Hong,
  • Ke Liu,
  • Deng Cai

摘要

This study investigated the impact of various state representation methods on vehicle decision-making at unsignalized intersections using deep reinforcement learning. We proposed a novel hybrid state representation framework that combines an attention mechanism with collision time prediction, substantially enhancing autonomous decision-making. Unlike traditional techniques, our approach fused complementary features to improve obstacle avoidance and incorporates an action masking module to optimize traffic efficiency. To demonstrate the practicality and efficiency of the proposed method, extensive simulation experiments were conducted, and the results validate its effectiveness.