Generation of Autonomous Vehicle Validation Scenarios Using Crash Data

Abstract

Autonomous vehicle (AV) technology is expected to make a transformative impact on mobility. As a step towards that direction, advanced driver-assistance systems (ADAS) and semi-autonomous features have been increasingly deployed in modern vehicles. Consequently, there have been crashes reported for these vehicles while operating in autonomous mode. In this paper, we describe a methodology to use the information from these crashes for the validation of autonomous vehicles, which is the most critical obstacle in their mass deployment. We use existing crash databases and provide a methodology to automatically convert them into test scenarios in simulation, using our scenario generation framework. The methodology assists in validation by recreating crashes and their cousin scenarios. It is also instrumental in forming the requirements the accident reports for autonomous vehicles.

Programmer

I do research on autonomous vehicle validation and verification as well as work part-time doing data visualizations and simulation programming.