This project is currently in the build phase. We are drafting guidelines to be published as white papers for public comment.
Autonomous Vehicle Vision
With the increasing deployment of AI-enabled autonomous vehicles, cyber assurance is crucial to protect against potential threats that could manipulate or disrupt the vehicle's perception of its surroundings and autonomous decision-making, compromising safety and trust in autonomous vehicle systems.
Developing cyber assurance to improve autonomous vehicles and accelerate their safe deployment.
The NCCoE is working on cyber assurance for autonomous vehicles by developing a public dataset and a testbed with difficult-to-handle and adversarial road/traffic conditions with the goal of improving autonomous vehicles and accelerating their safe deployment.
Status: Preparing Draft
Autonomous vehicles are increasingly prevalent in our day-to-day world. Cyber assurance—a comprehensive approach to ensure the security, integrity, and availability of systems and data—is needed to protect autonomous vehicle systems against threats. For autonomous vehicle vision systems, measuring the outputs from vehicle perception systems and motion forecasting algorithms, makes it possible to detect physically realizable cyber-attacks and measure the impact of those attacks on vehicle decisions and safety. In addition, by using generative artificial intelligence (AI) to identify rare failure modes—realistic scenarios in real-world conditions that autonomous vision systems fail to handle correctly—it is possible to accelerate testing and validation, and avoid costly, time-consuming reactive measures. Clear metrics to assess the cyber assurance of autonomous vehicles’ perception and decision-making will contribute toward building public trust and accelerating the safe deployment of autonomous vehicles.
The NCCoE is working with industry to develop a testbed to enable the discovery of machine learning (ML) and AI weaknesses that affect autonomous vehicle safety and is creating a public dataset with extraordinary events to help autonomous vehicle designers learn to develop safer and more secure autonomous vehicles. The NCCoE plans to publish white papers describing these efforts and providing more details on cyber assurance for autonomous vehicles.
For autonomous vehicle vision systems, measuring the outputs from vehicle perception systems and motion forecasting algorithms, makes it possible to detect physically realizable cyber-attacks and measure the impact of those attacks on vehicle decisions and safety.