Cutting-edge research from three top Virginia universities, led by the University of Virginia’s Homa Alemzadeh, is on a mission to revolutionize the safety of self-driving vehicles. With a substantial $926,737 grant from the National Science Foundation, this powerhouse team is dedicated to pinpointing and neutralizing potential computer failures in autonomous vehicle systems.
By harnessing this insight, they aim to fortify the resilience of the entire system and proactively eliminate safety risks. Alemzadeh, a trailblazing associate professor of electrical and computer engineering at UVA’s School of Engineering and Applied Science, is joined by the esteemed William & Mary professor of computer science, Evgenia Smirni, and the visionary lead investigator and George Mason University assistant professor of computer science, Lishan Yang.
The latest research reveals a concerning trend in the realm of autonomous vehicles: a significant number of “disengagements” occur due to incorrect decisions or delayed responses by machine learning-based AI, prompting shutdowns for safety reasons.
“We are particularly interested in studying disengagements and safety incidents due to transient hardware faults, temporary loss of network connection, or software errors,” she said.
Uncovering these elusive vulnerabilities is crucial, as these fleeting disruptions can be challenging to pinpoint and address. Despite being self-correcting in nature, they have the potential to create momentary chaos within the system, evading detection and posing serious risks.
It’s evident that the reliability and safety of self-driving vehicles hinge not only on physical components like sensors and brakes but also on the software “controller” responsible for executing autonomous decisions and machine learning functions. When transient faults manifest during critical operational phases, they have the capacity to permeate the system’s hardware and software layers, eluding existing safety protocols and precipitating hazardous situations.
“We aim to look at the end-to-end system — from input to output — to investigate the critical fault locations within the hardware and software, as well as the system contexts that lead to activation of faults and safety hazards,” Alemzadeh said.
The researchers are dedicated to enhancing controller and machine learning components to proactively prevent accidents. By employing cross-layer reliability analysis, they are strategically targeting critical faults within the intricate software code and hardware that underpin the controller and machine-learning models.
To put their solutions to the test in real time, the team will develop protection mechanisms that can automatically correct transient faults or mitigate unsafe vehicle operations by adjusting speed as needed based on when and where vulnerabilities are detected. These mechanisms will be selectively applied at different times and locations to ensure safety while optimizing efficiency.
Furthermore, the team will validate their solutions through closed-loop testing, simulating various weather, road, and traffic conditions to evaluate the performance of their safety features. Concurrently, they will simulate faults and errors to assess their impacts and the effectiveness of the solutions.
The three-year NSF project, “End-to-End Resilience in Autonomous Driving Systems: Strategic Vulnerability Assessment and Mitigation,” builds on the success of a previous collaboration led by Alemzadeh and her colleagues. Their groundbreaking work on “Toward Trustworthiness in Autonomous Vehicles,” funded by the Commonwealth Cyber Initiative Coastal Virginia regional node, has laid the foundation for this ambitious new project.
The Commonwealth Cyber Initiative, with its four regional nodes, represents a bold partnership between state higher education institutions, government, industry, and non-government organizations. Together, they are propelling Virginia to the forefront of cybersecurity research, innovation, and workforce development on a global scale.
