Spatially Clustered Autonomous Vehicle Malware: Producing New Urban Geographies of Inequity

Abstract

Malicious software (malware) is both a hurdle to autonomous vehicle (AV) adoption and a serious threat to AV occupant safety. Yet, to date, the topic of how subnational transportation decision makers should handle this cybersecurity threat has been underexplored. This paper describes how malware can spread through vehicle-to-X networks and infect AVs to a non-technical planning audience. It then explains how AV malware can cluster spatially in linguistic, socioeconomic, and political enclaves of American cities. Since AVs can identify new indicators of safety imperceptible to humans, malware clusters can produce new geographies of accessibility, mobility, economic, and environmental inequity across a city. In provisioning the inherently localized technical malware prevention tools available to them, subnational transportation planners have the capacity to reproduce or overcome historical and current transportation system inequities.

Guardauto: A Decentralized Runtime Protection System for Autonomous Driving

“…1, it also leads to a giant system consisting of tremendous processes interacting with each other to perform complex functions, including sensing, data processing, localization, route planning, motion control, trajectory control, etc., which makes it vulnerable to malware, software failures, or even a kernel panic [1], [2], [39], [40]. For example, malware can be remotely distributed into selfdriving cars [6], [23] to disrupt system execution or intrude the in-vehicle bus. Thus, it is urgent and necessary to decouple and isolate each functionality.…”

“…(2) implant malware to interfere the system execution, such as consuming available resources, sending fabricated control commands, or even disrupting critical services to cause system failures, which are frequently used in remote attacks [3], [6]. Software vulnerabilities may also trigger abnormal execution and lead to failures.…”

“…Besides, the current self-driving system consists of dozens of various functionalities, which may spawn hundreds of user processes interacting with each other through tremendous data exchange. As those functions are running together with other applications on top of the same kernel, any software failures or malware [6] could easily compromise the whole system or interfere the self-driving control. Moreover, it is also challenging to track the interaction among the driving functionalities, which makes it difficult to analyze and model the exact system behavior.…”