Autonomous vehicles (AVs) sit at the higher end of the both the complexity and safety-criticality spectrums of robotics, which means the regulators who must certify the vehicles for public use have a significant challenge. This talk will initially review current certification methods, which leverage decades of knowledge to create processes which are both highly effective and highly entrenched. However, these processes are not suitable for certifying AVs, due to the complexity of software systems needed to control an AV, and the range of conditions it needs to be able to handle safely (road layouts, road surroundings, other road users and their actions, weather conditions, etc) [1].
Despite the challenges, there is a feasible certification path building incrementally on existing test tools and methodologies. The critical pillars of this route are simulation, to increase test efficiency, and scenarios, to define relevant and human-understandable test cases [2,3]. Once a relatively simple scenario-based test methodology is established, it can evolve to include intelligent testing, likely based on some of the methods presented in other talks at this event. Further cultural shifts towards transparency could allow component-based or model-based test methods, opening the door applying more of the presented research. However, given the increasing presence of safety-critical AI technology in production vehicles, initial steps are vital to preserve the safety standards of regulatory approval.
[1] N. Kalra, S. M. Paddock: Driving to safety: How many miles of driving would it take to demonstrate autonomous vehicle reliability? RAND Corporation Technical Report (2016).
[2] Z. Saigol, A. Peters: Verifying Automated Driving Systems in Simulation: Framework and Challenges. In: 25th ITS World Congress, Copenhagen (2018).
[3] S. Hallerbach, Y. Xia, U. Eberle, F. Koester: Simulation-Based Identification of Critical Scenarios for Cooperative and Automated Vehicles. SAE Journal of Connected and Automated Vehicles (2018).
