Test Specification and Generation for Connected and Autonomous Vehicle in Virtual Road Environment

ABSTRACT: The trend of connected / autonomous features adds significant complexity to the traditional automotive systems. In order to improve driving safety and comfort, vehicles are expected to drive autonomously and/or to communicate with each other and infrastructures. Such complexity makes engineers harder to test correctness, performance or effectiveness of those driving features in the physical environment. In this talk, we introduce a virtual test framework that utilizes existing visualization engines (e.g., Unity3D, Unreal Engine or Prescan). In this test framework, a system component is integrated with a virtual vehicle that can be tested under a wide range of virtual road environments to overcome the limitation of the physical testing. In order to build such test environments, we introduce a formal way to specify geometric and behavioral aspects of the road environments using SMT constraints (Satisfiability Modulo Theories) and timed automata. We also introduce a systematic way to generate those road environments from the formal specification based on several test criteria. Finally, we show the applicability of the proposed road environment generation method using adaptive cruise control (an example of autonomous features) and right-turn pedestrian warning system (an example of connected features).

BIO: BaekGyu Kim earned B.S. and M.S. from Kyungpook National University in South Korea in 2007 and 2009, and earned Ph.D in computer science from University of Pennsylvania in 2015. His research interest is applying various formal techniques to build safety-critical real-time embedded systems according to the model-based development paradigm. His doctoral dissertation topic was to design model-based implementation framework to assure the safety of infusion pump systems (medical device) as a part of Generic Infusion Pump project. After joining Toyota InfoTechnology Center, he started applying those techniques to analyze correctness and effectiveness of automotive systems.

Hosted by Paul Bogdan