Finding Vulnerabilities of Autonomous Vehicle Stacks to Physical Adversaries
Autonomous Driving (AD) vehicles must interact and respond in real-time to multiple sensor signals, indicating how other autonomous robots, targets, and the environment behave near the ego vehicle. While autonomous vehicle (AV) developers tend to generate numerous test cases in simulations to detect problems, to the research team's best knowledge, they are not testing for malicious physical interactions from attackers, such as placing emergency cones in the hood of an AV or driving maneuvers that nearby human vehicle drivers or other AV manufacturers can create. For example, a hostile driving maneuver causing the victim vehicle to crash (while the malicious driver does not crash) can be identified by malicious actors and then spread and reproduced by multiple people worldwide, causing traffic accidents on vehicles with vulnerable AD stacks. Recently, National Center for Transportation Cybersecurity and Resiliency (TraCR) members of University of California, Santa Cruz (UCSC) and Purdue have introduced two frameworks to explore the practicability of adversarial physical conditions in real-world environments. They focused on adversarial driving maneuvers, a new class of physical attack against AD software. Here, the attacker aims to find a (plausible) trajectory near the victim's vehicle to cause it to behave unintendedly, such as crashing or driving off the road. The frameworks proposed by UCSC and Purdue differ in their assumptions about the attacker and the target AV software components. However, both provide an overview of the challenges, a means of discovering adversarial driving maneuvers in practice, and potential solutions to defend against them. While both frameworks have been shown, to some extent, to be effective in discovering adversarial driving maneuvers against a variety of AD software, the research on adversarial driving maneuvers is still in its early stages. In this proposal, the team will study the weaknesses and strengths of both frameworks. Guided by their findings, the team will explore creating a unified framework leveraging the best ideas from each university and explore rigorous measures of adversarial maneuvers for building a safe and secure AD software stack.
Language
- English
Project
- Status: Active
- Funding: $251852
-
Contract Numbers:
69A3552344812
69A3552348317
-
Sponsor Organizations:
Office of the Assistant Secretary for Research and Technology
University Transportation Centers Program
Department of Transportation
Washington, DC United States 20590 1040 South River Road
West Lafayette, IN United States 47907University of California, Santa Cruz
1156 High Street, Mail Stop SOE2
Santa Cruz, California United Kingdom 95064 -
Managing Organizations:
National Center for Transportation Cybersecurity and Resiliency
1 Research Dr
Greenville, South Carolina United States 29607 1040 South River Road
West Lafayette, IN United States 47907 -
Project Managers:
Chowdhury, Mashrur
-
Performing Organizations:
1040 South River Road
West Lafayette, IN United States 47907University of California, Santa Cruz
1156 High Street, Mail Stop SOE2
Santa Cruz, California United Kingdom 95064 -
Principal Investigators:
Celik, Berkay
Cardenas, Alvaro
Fremont, Daniel
Ukkusuri, Satish
- Start Date: 20240101
- Expected Completion Date: 20241231
- Actual Completion Date: 0
- USDOT Program: University Transportation Centers
Subject/Index Terms
- TRT Terms: Autonomous vehicles; Computer security; Crimes; Driving behavior; Physical phenomena; Risk assessment; Safety
- Subject Areas: Planning and Forecasting; Safety and Human Factors; Security and Emergencies; Transportation (General); Vehicles and Equipment;
Filing Info
- Accession Number: 01906983
- Record Type: Research project
- Source Agency: National Center for Transportation Cybersecurity and Resiliency (TraCR)
- Contract Numbers: 69A3552344812, 69A3552348317
- Files: UTC, RIP
- Created Date: Feb 5 2024 3:58PM