Scholarlink

<p>This paper presents a framework to improve the fidelity of vehicle-in-the-loop (VIL) simulation by incorporating realistic LiDAR modeling and environment reconstruction. While VIL inherently captures real vehicle dynamics, conventional setups still suffer from a sim-to-real gap due to simplified sensor emulation and generic virtual environments. To address these limitations, we replicate LiDAR beam patterns and packet structures, and construct realistic environments from point cloud maps using Poisson surface reconstruction. Five case configurations were evaluated under Euro NCAP car-to-car scenarios to isolate the incremental contributions of sensor and environment realism. Fidelity was quantified using normalized RMSE and Pearson correlation computed from six key signals. Results show that environment realism provides the largest fidelity gains, particularly in dynamic crossing and braking scenarios, by stabilizing localization through accurate geometric reconstruction. Realistic LiDAR offers supplementary fine-tuning that further improves perception accuracy. The fully integrated configuration achieved relative fidelity scores approaching 100, closely matching ground-truth behavior. These findings demonstrate that maximizing realism in sensor and environment domains is essential for reliable and cost-efficient validation of autonomous driving systems.</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

VIL Simulation Fidelity Improvement Through Realistic Environment and Sensor Modeling

  • 期刊论文

  • 发表日期：

    2026年7月7日

查看全文
  • J. Hong,
  • M. Kim,
  • H. Shin,
  • W. Sung

摘要

This paper presents a framework to improve the fidelity of vehicle-in-the-loop (VIL) simulation by incorporating realistic LiDAR modeling and environment reconstruction. While VIL inherently captures real vehicle dynamics, conventional setups still suffer from a sim-to-real gap due to simplified sensor emulation and generic virtual environments. To address these limitations, we replicate LiDAR beam patterns and packet structures, and construct realistic environments from point cloud maps using Poisson surface reconstruction. Five case configurations were evaluated under Euro NCAP car-to-car scenarios to isolate the incremental contributions of sensor and environment realism. Fidelity was quantified using normalized RMSE and Pearson correlation computed from six key signals. Results show that environment realism provides the largest fidelity gains, particularly in dynamic crossing and braking scenarios, by stabilizing localization through accurate geometric reconstruction. Realistic LiDAR offers supplementary fine-tuning that further improves perception accuracy. The fully integrated configuration achieved relative fidelity scores approaching 100, closely matching ground-truth behavior. These findings demonstrate that maximizing realism in sensor and environment domains is essential for reliable and cost-efficient validation of autonomous driving systems.