Time-efficient RobUSt verification of auTonomuous vehicles - theory and Methods (TRUST-ME)

Purpose and goal

Verification of safety and reliability is expected to be one of the biggest challenges for commercialization of autonomous vehicles. The purpose of the verification is to quantitatively show that the vehicle solves all situations as well, or better than the human driver. The project´s goal is to develop and validate autonomous vehicle verification methods. The developed method has reached the goal of estimating the safety of autonomous vehicles from field data. The method also systematically identifies critical situations that determine the performance of autonomous vehicles.

Expected results and effects

This project has developed a method that can be used to validate the safety of a vehicle’s driving. Data captured during real traffic driving is used to evaluate the closeness to a collision, which is extrapolated into a collision frequency using extreme value theory (EVT). Different types of measures for the closeness to a collision, as well as methods to correctly fit the EVT model to the data, has been evaluated. Based on these results, the usage of EVT for safety validation looks promising.

Planned approach and implementation

The project has been conducted primarily through an industrial PhD candidate initially employed by Volvo Cars and later by Zenuity. Chalmers has been in charge of academic supervision and Volvo Cars / Zenuity has focused on linking the project´s work to internal and external development projects for autonomous vehicles (e.g. Drive Me). The project has had the principle of publishing all method development in scientific conferences and journals.