Autonomous vehicles can be imperfect – as long as they are resilient

Researchers at three top Virginia universities, including Homa Alemzadeh of the University of Virginia, are aiming to reduce the risks associated with autonomous vehicles by overcoming inevitable computer failures with good engineering.

The trio will share a $926,737 National Science Foundation award to determine when and where autonomous vehicle systems are most at risk of safety-critical failures. They plan to use this knowledge to design ways to effectively mitigate potential security threats and increase overall system resilience.

Alemzadeh, an associate professor of electrical and computer engineering in the UVA School of Engineering and Applied Sciences, is collaborating with William & Mary professor of computer science Evgenia Smirni and assistant professor of computer science at George Mason University Lishan Yang, the principal investigator.

Alemzadeh said recent research shows that a significant portion of “shutdowns,” in which an autonomous system is turned off by the system or driver for safety reasons, occur when machine learning-based AI makes poor or timely decisions.

“We are particularly interested in investigating disconnections and security incidents caused by transient hardware failures, temporary loss of network connection or software errors,” she said.

Hunting for elusive vulnerabilities

These often self-correcting events may cause only momentary disruptions and then disappear, making them difficult to find and diagnose.

When it comes to the safety and reliability of autonomous vehicles, they depend as much on the software “controller” that makes and implements decisions about autonomous operation and machine learning components as they do on physical parts such as sensors and brakes.

If transient faults are activated at a critical point in operation, they can propagate through the system’s hardware and software layers, bypass existing security controls and create threats.

“We want to look at the entire system – from input to output – to investigate the critical locations of faults in hardware and software, as well as the system contexts that lead to activation of faults and security threats,” Alemzadeh said.

Scientists will focus on improving controllers and machine learning components to prevent accidents. Using cross-layer reliability analysis, they plan to strategically locate the most critical bugs in the vast, complex software code and hardware that underlie controllers and machine learning models.

Real-time test drive solutions

Based on when and where they find vulnerabilities, the team will design safety mechanisms—such as automatically correcting transient faults or mitigating unsafe vehicle operations through automatic deceleration—that can be applied selectively at different times and locations to ensure safety while maximizing efficiency.

The team will validate its solutions through closed-loop testing, during which the autonomous system and its safety functions will be tested in real time using driving simulations in changing weather, road and traffic conditions. They will simultaneously simulate faults and errors to assess their impact and the performance of their solutions.

The three-year NSF project, End-to-End Resilience of Autonomous Driving Systems: Strategically Assessing and Mitigating Vulnerabilities, is a continuation of previous collaborations by Alemzadeh and her colleagues. The “Toward Autonomous Vehicle Credibility” project, funded by the Commonwealth Cyber ​​Initiative Coastal Virginia Regional Node, laid the foundation for the current project.

The Commonwealth Cyber ​​Initiative, with four regional nodes, is a coalition of state higher education institutions that works with government, industry and non-governmental organizations to make Virginia a global leader in cybersecurity through research, innovation and workforce development.

Alemzadeh is also a member of UVA Engineering’s Link Lab, a multidisciplinary research and education center for cyberphysical systems, and holds a courtesy appointment in computer science.

Provided by the University of Virginia