Safety evaluation is crucial for the mass deployment of highly automated vehicles (HAVs). This paper presents a procedure for conducting behavior competence testing for HAVs. First, we describe an efficient test case generation scheme for cut-in and unprotected left turn scenario. We then design the algorithm for synchronizing the primary other vehicle with the vehicle under test based on optimal control. Then, scoring criteria for quantifying HAV performances is presented. Finally, we implement the testing procedure in both simulations and real world. Accurate and repeatable testing is achieved, and quantitative evaluation results are acquired for a baseline HAV.

View full abstract

Naturalistic driving studies (NDS) are an increasingly popular method to research driving behavior. They often result in large amounts of data varying in source and format (videos, spatial, and time-series data). Traditional data processing systems and analytical methods are not equipped to handle the large influx of data, often ranging from terabytes to petabytes. Previously, big data analytics platforms have been designed to address specific use cases of intelligent transport systems such as traffic flow prediction, transportation planning, and traffic safety. Similarly, there is a need for robust data systems for storing, mining, visualizing, and analyzing big naturalistic data. This paper presents a comprehensive cloud-based AI platform, Deep Insight, designed for data management, modeling, and enhanced annotations of naturalistic driving data. The platform capitalizes on Amazon Web Services, hosting a repository of public and privately collected NDS datasets with tool integration for data annotation and machine learning modeling that permits data analysis and inference. This end-to-end framework provides effective and reliable tools for storing, processing, annotating, and modeling NDS datasets. Additionally, the platform hosts a metric dashboard for benchmarking and displaying the performance of diverse analytical models using a standard dataset. The authors present a case study classifying a driver's head movement to demonstrate this framework’s workflow using Deep Insight and integrated tools. This cloud-based platform offers a wide range of cost, access, scalability, and security benefits, supporting goals to create a one-stop, standardized destination for analyzing naturalistic driving data and studying driver behavior.

View full abstract

In this study, a computer simulation technique is developed for numerically predicting the residual stress relaxation caused by bending-straightening in quenching-induction-hardened axle shafts. Both the strength and residual stress distributions inside the shaft before bending-straightening are taken into account in the simulation model. The calculated residual stress distribution is compared with that measured using the contour method to discuss the effectiveness of the developed technique. In addition, the generation of plastic strain due to bending-straightening is considered thorough a theoretical analysis based on material mechanics in addition to the numerical results obtained using the developed simulation technique.

View full abstract