Saeedeh Lohrasbi, PhD student in Systems Design Engineering, has carved a notable path in the realm of autonomous vehicle (AV) technology. Joining the Automation and Intelligent Systems Group as a research assistant in 2017, Saeedeh has made significant strides in integrating advanced machine learning techniques, particularly reinforcement learning (RL), into AV systems, with a steadfast focus on enhancing safety and security.
Her academic pursuit began with the development of a robust platform aimed at training RL-based controllers tailored for autonomous driving scenarios. Recognizing the critical importance of safety in AV operations, Saeedeh implemented safe RL techniques to ensure the controllers could navigate complex environments while adhering to stringent safety protocols.
Central to her research was the creation of a highly realistic simulation environment. This environment was meticulously designed to bridge the gap between virtual simulations and real-world implementation. It encompassed intricate details such as vehicle dynamics, varied road and traffic conditions, and nuanced agent interactions. By integrating multiple software tools, Saeedeh addressed existing limitations in deep reinforcement learning (DRL) studies, which too often rely on models that does not capture the full complexity of real-world AV operations.
Transitioning from foundational work in RL-based control systems, Saeedeh turned his attention to adversarial attacks, which exploit weaknesses in machine learning algorithms that can deceive the DL and RL model without being noticed. These attacks pose a significant threat to AI-driven AVs, potentially compromising their functionality and safety. Saeedeh devised a novel adversarial attack algorithm specifically targeting RL-based AVs. Her algorithm demonstrated superior efficacy compared to conventional baseline attacks documented in current literature. Currently, she is immersed in developing robust defense mechanisms to mitigate the impact of such attacks, aiming to fortify the resilience of future autonomous vehicle systems against adversarial threats.
The interdisciplinary approach underscores Saeedeh’s commitment to advancing the frontiers of autonomous vehicle technology. Her research not only pushes boundaries in autonomous systems but also contributes significantly to enhancing the safety, security, and reliability of AI-driven vehicles on our roads. As she continues to innovate and collaborate with industry and academic partners alike, Saeedeh remains dedicated to shaping the future of autonomous mobility.