Proceedings of the
European Safety and Reliability Conference (ESREL2026)
14 – 19 June 2026, Braga, Portugal
Towards safe deployment of autonomous vehicles: responsibilities, hazards, and metrics identification based on STPA methodology
1B. John Garrick Institute for the Risk Sciences, University of California, Los Angeles, USA.
2Center of Excellence on New Mobility and Automated Vehicles, University of California, Los Angeles, USA.
ABSTRACT
The autonomous vehicle (AV) industry has significantly evolved since the initiation of the AV testing program by the California Department of Motor Vehicles (DMV) in 2012. As commercial deployments for passenger service transport expand to multiple cities, key challenges remain to comprehensively assess the AV fleet's safety performance and their impact on traffic systems. In the U.S., AV developers primarily report limited and siloed incident metrics that offer little insight into how risks may evolve as AV fleets scale. This limits the ability of regulatory entities to make informed decisions about AV integration within complex urban traffic systems. Therefore, this work proposes a model-based methodology to elicit operational safety-related metrics as part of an ongoing effort to develop risk-informed tools enhancing the capacity of local jurisdictions to assess an AV fleet's safety performance. This work leverages System-Theoretic Process Analysis (STPA) to identify and model safetycritical interactions between AV fleets and other road users across deployment stages, aiming to identify hazards arising from unsafe control actions (UCAs) that could result in system-level losses beyond traditional crash metrics, such as delayed emergency response. The STPA-based analysis accounts for interacting entities (e.g., AVs, humandriven vehicles), responsibilities of the entities (e.g., preventive or mitigative controls), control actions (e.g., remote operators send commands to AVs), and feedback mechanisms (e.g., AVs send environmental conditions back to remote operators). UCAs are identified through analysing control actions in the operational context of AV fleets, supported by available operational and incident data, and assessed for public-reporting feasibility. This work presents the functional control and interaction model of a generic AV fleet, a structured catalogue of critical UCAs, associated hazard scenarios, and corresponding performance metrics linked to system safety objectives.
Keywords: Autonomous vehicles, testing and deployment, safety analysis, STPA.
