세미나

(Title) Scenario Specification and Generation for Testing Autonomous Driving Systems in Virtual Environment

(Abstract)
Along with the trend of Digital Twin, the virtual test is a widely adopted concept to validate the safety and correctness of autonomous vehicle software in the virtual world to overcome the major limitations of the physical test, such as scalability, reproducibility, and repeatability.
In this talk, we introduce a test framework to systematically create diverse driving scenarios in virtual worlds to improve the virtual test quality. A driving scenario consists of static objects (e.g., roadways or side objects) and dynamic objects (e.g., pedestrians or vehicles); their characteristics are modeled as a set of parameters at two different abstraction levels, and we call them abstract and concrete scenarios.
In the abstract scenario, the high- level semantics of such objects are expressed as combinations of parameter intervals, for example, the number or distance of curves in a roadway or possible lanes that a vehicle can drive at a certain time interval.
On the other hand, in concrete scenario, the valuations of the parameters are automatically determined, guaranteeing the semantics of the abstract scenario using well-known solvers such as SMT (Satisfiability Modulo Theories) and MILP (Mixed-Integer Linear Programming). We present case studies to show (i) how existing industrial test standards, such as ISO test standards and ASAM OpenSCENARIO, can be formalized using our framework and (ii) how virtual environments are auto-generated for testing adaptive cruise control, vision-based lane-keeping assistance and right-turn pedestrian warning.

Bio (Korean)
BaekGyu Kim is an Assistant Professor at Department of Electrical Engineering and Computer Science at DGIST. He received a Ph.D. degree in Computer Science from University of Pennsylvania in 2015, and B.S and M.S degrees in Electrical Engineering and Computer Science from Kyungpook National University in 2007 and 2009. Before joining DGIST, he was a principal researcher at Toyota Motor North America R&D –  InfoTech Labs. in Mountain View, CA, USA, where he worked for about 6 years. His primary research area is to guarantee software quality assurance in Cyber Physical Systems and Internet of Things. He published more than 70 international papers and 60 international patents. He received the top inventor award from Toyota InfoTechnology Center, U.S.A. (2017), and his collaborative work received SAE Vincent Bendix Automotive Electronics Engineering Award (Best paper award, 2019).