Section: New Results

Statistical Model-Checking for Autonomous Vehicle Safety Validation

Published in [20].

We present an application of statistical model-checking to the verification of an autonomous vehicle controller. Our goal is to check safety properties in various traffic situations. More specifically, we focus on a traffic jam situation.

The controller is specified by a C++ program. Using sensors, it registers positions and velocities of nearby vehicles and modifies the position and velocity of the controlled vehicle to avoid collisions. We model the environment using a stochastic high level Petri net, where random behaviors of other vehicles can be described. We use HASL, a quantitative variant of linear temporal logic, to express the desired properties. A large family of performance indicators can be specified in HASL and we target in particular the expectation of travelled distance or the collision probability.

We evaluate the properties of this model using COSMOS1. This simulation tool implements numerous statistical techniques such as sequential hypothesis testing and most confidence range computation methods. Its efficiency allowed us to conduct several experiments with success.