Privacy-Preserving Data Collection Frameworks for Autonomous Vehicle Telemetry

Abstract

Despite clear demand for this kind of real-world experience data, it is difficult to see how to sustainably and widely collect and use it. The recent debates about the role of third-party data and direct vehicle-to-government data reporting in new federal guidance for vehicle safety compliance and testing provide a leading-edge illustration of just how expensive and contentious it might be to create any kind of centralized collection infrastructure for this kind of data. In this paper, we present several privacy-preserving data collection frameworks that collect the semantics of physical-world events and aggregate the probe data inside the autonomous vehicle before transmitting them in raw form, after appropriate filtering and reduction, to a cloud-based data broker. We discuss the tradeoffs and limitations of this approach. Our protocols contribute to the formation of a marketplace to inform law enforcement, regulators, insurers, and the public about the availability, format, and prices of different types of data that are desired to produce the kind of transparency and accountability that society will expect from those that will be allowed to purchase, build, operate, and benefit from autonomous vehicle technology.

Autonomous vehicle technology holds great promise to revolutionize personal and commercial transportation and make mobility more convenient, efficient, and accessible. This promise, however, is accompanied by a host of challenges with respect to public safety, consumer trust, integration with legacy modes, insurance and liability, and cybersecurity, to name just a few. A recurring theme across many such challenges is the importance of robust, empirically-based evidence about how autonomous vehicles actually perform in the real world in order to enable law enforcement, regulators, insurers, and consumers to trust that vehicles will work as intended in a wide variety of conditions. For example, in the unfortunate event of a crash or other property damage, incident investigators will seek to understand whether the vehicle was functioning properly at the time of the event, and incident investigators, law enforcement, and autonomous vehicle manufacturers may benefit from real-world data about incidents to understand the nature of public risk and to shape responses.