Accelerated Evaluation of Automated Vehicles Safety in Lane-Change Scenarios Based on Importance Sampling Techniques

Zhao, Ding; Lam, Henry; Peng, Huei; Bao, Shan; LeBlanc, David J.; Nobukawa, Kazutoshi; Pan, Christopher S. · 2016 · OpenAlex-citations

DOI: 10.1109/tits.2016.2582208

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the challenge of efficiently evaluating the safety of Automated Vehicles (AVs) before deployment. Traditional Naturalistic Field Operational Tests (N-FOTs) are prohibitively expensive and time-consuming because safety-critical events, such as crashes, are rare in naturalistic driving. The authors propose an accelerated evaluation method using Importance Sampling (IS) techniques to simulate risky scenarios more frequently while maintaining statistical accuracy. The study specifically targets frontal collisions caused by unsafe lane changes (cut-ins) by human-controlled vehicles, aiming to reduce the development and validation time for AVs. The methodology utilizes data from the University of Michigan Safety Pilot Model Deployment (SPMD) program, which contains over 400,000 recorded lane change events. The authors first model human driver behaviors by analyzing key variables: the velocity of the lane-changing vehicle ($v_L$), the range ($R_L$), and the Time To Collision ($TTC_L$). They fit these variables to statistical distributions, such as Pareto and Exponential distributions, to capture the tail events representing risky maneuvers. To accelerate the evaluation, the authors employ the Cross Entropy (CE) method to recursively search for optimal parameters that skew these probability density functions. This "morphing" of statistics generates a higher frequency of dangerous cut-in scenarios in simulations. The likelihood ratio is then used to correct the bias introduced by the skewed sampling, allowing for an accurate estimation of real-world crash rates. The results demonstrate that the proposed accelerated evaluation technique achieves an acceleration rate of approximately 2,000 to 20,000. This implies that driving 1,000 miles in the accelerated simulation exposes the AV to challenging scenarios equivalent to 2 to 20 million miles of real-world driving. The method successfully estimates the frequencies of conflicts, crashes, and injuries for the modeled AV. By using the CE method to automatically tune the importance sampling parameters, the approach ensures high accuracy and reliability in estimating rare event probabilities, overcoming the limitations of crude Monte Carlo simulations which require infinite samples for rare events. The significance of this work lies in providing a practical, efficient framework for AV safety validation. It allows developers to objectively identify critical test scenarios and estimate social benefits without the massive costs associated with N-FOTs. The technique is applicable to various evaluation platforms, including driving simulators, on-track tests, and hardware-in-the-loop systems. By enabling the rapid assessment of AV interactions with human drivers, this method supports the safe and timely deployment of automated vehicle technologies.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• lane changing

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Empirical Findings: crash risk outcomes
• Theoretical Contribution: computational model