Test Specification and Generation for Connected and Autonomous Vehicle in Virtual Environments

Abstract

The trend of connected/autonomous features adds significant complexity to the traditional automotive systems to improve driving safety and comfort. Engineers are facing significant challenges in designing test environments that are more complex than ever. We propose a test framework that allows one to automatically generate various virtual road environments from the path and behavior specifications. The path specification intends to characterize geometric paths that an environmental object (e.g., a roadway or a pedestrian) needs to be visualized or move over. We characterize this aspect in the form of constraints of 3-Dimensional (3D) coordinates. Then, we introduce a test coverage, called an area coverage, to quantify the quality of the generated paths in terms of how diverse of an area the generated paths can cover. We propose an algorithm that automatically generates such paths using an SMT (Satisfiability Modulo Theories) solver. However, the behavioral specification intends to characterize how an environmental object changes its mode over time by interacting with other objects (e.g., a pedestrian waits for a signal or starts crossing). We characterize this aspect in the form of timed automata. Then, we introduce a test coverage, called an edge/location coverage, to quantify the quality of the generated mode changes in terms of how many modes or transitions are visited. We propose a method that automatically generates many different mode changes using a model-checking. To demonstrate the test framework, we developed the right-turn pedestrian warning system in intersection scenarios and generated many different types of pedestrian paths and behaviors to analyze the effectiveness of the system.