Scholarlink

<p>Non-line-of-sight occlusions at superhighway speeds create perceptual blind spots that challenge autonomous collision avoidance. Although vehicle-to-everything (V2X) communication extends perception through roadside sensing, sensor noise and random communication delay degrade conventional filtering-based estimation, while Monte Carlo risk evaluation is often too expensive for real-time use. This paper proposes V2X-CDU-CRA, a cooperative collision-risk assessment framework for high-speed cut-in scenarios under V2X dual uncertainty. A Constant Turn Rate and Acceleration motion model is integrated into an asynchronous factor-graph estimator, where delayed V2X measurements are inserted at their generation timestamps and optimized by iSAM2 to avoid forward-extrapolation error. An Analytical Probabilistic Risk Field is then derived to compute collision probability in closed form using the Gaussian error function, and a rotated variant is introduced to improve accuracy under sideslip. In CARLA-based simulations with Gamma-distributed delays up to 800&#xa0;ms, the proposed method reduces lateral RMSE by up to 91.6% relative to EKF-MC at 160&#xa0;km/h, achieves a risk response time of 1.51&#xa0;s, yields an expected calibration error of 0.00298, and runs in 5.1&#xa0;ms per frame.</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

Cooperative Collision Risk Assessment for Superhighway Cut-In Scenarios Under V2X Dual Uncertainty

  • 期刊论文

  • 发表日期：

    2026年6月22日

查看全文
  • Yongming He,
  • Xinran Li,
  • Yangpeng Lu,
  • Huiyang Liu

摘要

Non-line-of-sight occlusions at superhighway speeds create perceptual blind spots that challenge autonomous collision avoidance. Although vehicle-to-everything (V2X) communication extends perception through roadside sensing, sensor noise and random communication delay degrade conventional filtering-based estimation, while Monte Carlo risk evaluation is often too expensive for real-time use. This paper proposes V2X-CDU-CRA, a cooperative collision-risk assessment framework for high-speed cut-in scenarios under V2X dual uncertainty. A Constant Turn Rate and Acceleration motion model is integrated into an asynchronous factor-graph estimator, where delayed V2X measurements are inserted at their generation timestamps and optimized by iSAM2 to avoid forward-extrapolation error. An Analytical Probabilistic Risk Field is then derived to compute collision probability in closed form using the Gaussian error function, and a rotated variant is introduced to improve accuracy under sideslip. In CARLA-based simulations with Gamma-distributed delays up to 800 ms, the proposed method reduces lateral RMSE by up to 91.6% relative to EKF-MC at 160 km/h, achieves a risk response time of 1.51 s, yields an expected calibration error of 0.00298, and runs in 5.1 ms per frame.