The Computing Community Consortium (CCC) is pleased to share that council member Weisong Shi was awarded two grants from the National Science Foundation (NSF) to further autonomous research at the University of Delaware.
Here’s more about his research:
Machine Learning (ML) has become a driving force in advancing many critical domains, particularly in autonomous driving. The growing power of ML techniques enables autonomous vehicles (AVs) to perceive their surroundings accurately, make real-time decisions, and operate reliably without human input. Given the safety-critical nature of AVs, ensuring the end-to-end safety of their learning systems is paramount.
However, Shi’s research indicates that potential safety issues can arise from various sources, including the model, hardware, and system. For example, ML models trained on typical driving scenarios might perform poorly in rare or unique situations, leading to unsafe decisions. Hardware platforms running these models can also experience faults and errors. When multiple ML models run concurrently, the real-time operating system of AVs may fail to deliver decisions promptly, which could result in catastrophic outcomes.
Since joining the University of Delaware in Fall 2022, Shi has been developing a dynamic testbed called D-STAR for connected and autonomous driving on the Science, Technology and Advanced Research (STAR) campus. D-STAR features three autonomous vehicles, a private 5G network, C-V2X enabled intersections, an Indoor Connected and Autonomous Testbed (ICAT), and a digital twin of the entire system hosted in the AWS Cloud.
Shi credits these two research grants as a direct result of the D-STAR testbed, enabling his team to create safe and reliable computing systems for autonomous vehicles.
“In 2023, we observed significant variability in how deep neural networks (DNN) were applied to perception tasks in autonomous driving, posing a serious risk to vehicle safety,” Shi said. “Thanks to the support from the National Science Foundation, we are now able to tackle this challenging problem and share our discoveries with the broader community.”
Shi’s ultimate goal is to develop a predictable operating system for autonomous vehicles.
