Adversarial booking attack for autonomous on-demand mobility services
Primary Investigator:
Satish Ukkusuri
Zengxiang Lei, Satish V. Ukkusuri
Abstract
On-demand mobility services, such as Uber and Lyft, are at the forefront of transforming transportation operations by providing online vehicle scheduling and routing. Recently, fully controlled fleets have also been realized through autonomous driving. Despite the clear benefits of introducing real-time controls in transportation operations, the vulnerabilities and associated risks are largely understudied. In this study, we investigate a new attack model named "adversarial booking attacks" to measure the risks inherent in the core operation of on-demand mobility services--the request-vehicle matching process. The attack involves malicious entities who manipulate multiple accounts to generate purposefully crafted trip requests, aiming to disrupt service operations. We formulate the attack into an optimization problem with objectives to reduce the matching pairs and maximize induced traffic in a specific region. Using real-world ride-hailing trip records from New York City and traffic simulator SUMO, we explore the potential large-scale impact of adversarial booking attacks under various demand scenarios and attack strategies.