Development of Evaluation Approaches and the Certificate System for Automated Vehicles Based on the Accelerated Evaluation
Henry Lam, PhD
Assistant Professor, Industrial & Operations Engineering
May 2016 – December 2017
David LeBlanc, PhD
Associate Research Scientist, U-M Transportation Research Institute
Ding Zhao, PhD
Research Fellow, U-M Transportation Research Institute
To develop test approaches to evaluate the safety of Automated Vehicles (AVs) that reduces the test duration from a few months to a few days without sacrificing accuracy, and to identify possible failure modes on perception errors and misoperations in how to respond.
Tests are developed in two layers: scenarios and variations. Test scenarios are extracted from naturalistic driving data, crash data and expert knowledge. For each scenario, test variations are developed based on naturalistic driving data collected in the Connected Ann Arbor project. Statistical tools will be developed, based on the Importance Sampling approach, to measure safety impacts efficiently by reducing the safe and boring events while emphasizing the safety-critical conditions.
The findings in this project provide an approach to evaluate the safety level of an automated vehicle
- We adopted the GMM as a statistical representation of the uncertainty in the naturalistic driving environment
- We developed new methods for accelerating the test procedure of AVs. The new methods integrate novel techniques in statistical learning, optimization, experimental design, ect.
- The proposed methods have potential to largely improve the efficiency of an AV safety evaluation procedure