Back to search

FRINATEK-Fri prosj.st. mat.,naturv.,tek

Theory, methods, and tools for testing autonomous systems

Alternative title: Theory, methods, and tools for testing autonomous systems

Awarded: NOK 8.0 mill.

Autonomous systems (AS) are emerging technologies that have the potential to impact many areas of human life. In T3AS, we are exploring autonomous cars (AC) as one popular type of AS. ACs are well known for being vulnerable to adversarial attacks that can compromise the safety of the car and pose a danger to other road users. To defend against adversarial attacks in ACs, we have developed a two-step methodology for testing ACs in a multi-agent environment, with multiple cars involved in a driving scenario. The methodology consists of (i) training an adversarial driving agent that can find failure states in ACs, (ii) reward modeling to create a behavior representation that allows finding likely uncertain behavior in ACs, and (iii) improving the robustness of ACs by retraining them with effective adversarial inputs.

Autonomous systems are emerging technologies that are impacting a range of industries and many areas of human life nowadays. Driverless cars have started appearing on public roads, and collaborative robots have started working with human workers on the factory floor. Autonomous technologies offer a significant opportunity to enhance economy and society, but they may also cause fatal harm if they malfunction. There are great open challenges of testing these autonomous systems, to prevent their malfunctioning, and to ensure their safe and fault-free behavior. T3AS project will develop a novel theoretical foundation based on artificial intelligence, with a set of methods and tools, for addressing the great challenges of testing autonomous systems, to make them dependable and safe for their users and environment.

Publications from Cristin

No publications found

No publications found

No publications found

No publications found

Funding scheme:

FRINATEK-Fri prosj.st. mat.,naturv.,tek