The chair of the defence will moderate the disputation.
Ex auditorio questions: the chair of the defence will invite the attending audience to ask ex auditorio questions.
Trial lecture
"'Security assurance for automotive systems'"
Time: September 27, 2024. 11.15 AM, Kristen Nygaards sal (5370), Ole Johan Dahls hus / zoom.
Main research findings
-
Autonomous cars, using advanced technologies and deep learning models, aim to enhance road safety and reduce traffic congestion. However, due to their novelty and sensitivity to environmental factors, these vehicles require thorough testing to ensure reliability and safety before widespread adaptation. This thesis investigates the safety and robustness of autonomous cars in complex multi-agent environments and when trained using end-to-end deep reinforcement learning algorithms.
-
Three methodologies are proposed:
-
1. We developed MAD-ARL, a method that trains adversarial driving agents to identify failure states in autonomous cars. This is followed by retraining to enhance robustness against potential adversarial attacks.
-
2. We constructed a benchmarking framework to compare deep reinforcement learning algorithms in autonomous driving agents. This framework enables systematic evaluation in both single-agent and multi-agent scenarios.
-
3. We created ReMAV, an intelligent reward modeling-based testing framework. It generates a behavior representation to analyze uncertain driving states, focusing on minimal perturbation attacks where driving decisions are less confident.
-
-
These methodologies test the robustness of autonomous cars with and without adversarial attacks, demonstrating effectiveness in finding failure events such as increased collisions and offroad steering errors. The thesis aims to improve the safety and resilience of autonomous cars, making them more reliable and better equipped to handle potential threats in real-world driving scenarios.
Adjudication committee:
- Senior lecturer Markus Borg, Lund University, Sweden
- Associate Professor Narges Khakpour, Newcastle University, UK
- Professor Tor Skeie, Institute of Informatics, University of Oslo, Norway
Supervisors:
- Senior Research Scientist Dusica Marijan, Simula Research Laboratory
- Chief Research Scientist Arnaud Gotlieb, Simula Research Laboratory
- Professor Are Magnus Bruaseth, Simula Research Laboratory
Chair of defence:
Professor Stephan Oepen
Contact information at Department: Mozhdeh Sheibani