Topic: Measuring Stability and Robustness of Autonomous Driving Perception Systems Under Dynamic Conditions

Speaker: Liang Shi

Date: Thursday, August 13, 2026, 10:00 AM EST/ 7:00 AM PST/ 4:00 PM CEST

Abstract

Reliable perception is essential for autonomous driving systems, especially when vehicles operate under dynamic and adverse environmental conditions. Conventional evaluation metrics such as average precision, IoU, and F1 score are useful for static frame-level assessment, but they often fail to capture how perception reliability changes with distance, uncertainty, weather, and illumination. In this webinar, I will introduce Perception Characteristics Distance (PCD), an uncertainty-aware metric designed to quantify the maximum distance at which a perception system can consistently produce reliable detections under a specified decision rule. I will also discuss the SensorRainFall dataset, collected on the Virginia Smart Roads under controlled clear, rainy, daylight, night, and streetlight conditions, with manually annotated bounding boxes and segmentation masks for vehicle and pedestrian targets. Using benchmarks, I will show how PCD reveals important robustness characteristics that traditional metrics may overlook, and how this framework can support safer evaluation of perception systems for ADAS and autonomous driving applications.

Speaker Bio

Liang Shi is a Research Associate at the Virginia Tech Transportation Institute (VTTI), working in computer vision, machine learning, and transportation safety. He received his Ph.D. in Statistics from Virginia Tech and also holds a master’s degree in Computer Science. His research focuses on AI perception, vision-language models, naturalistic driving data, autonomous driving safety, and the evaluation of machine learning systems under real-world and safety-critical conditions. His recent work includes research on perception robustness for autonomous driving, synthetic and multimodal benchmarks for driving safety-critical events, and vision-language understanding of traffic safety scenarios. He is also an Adjunct Professor in the Department of Statistics at Virginia Tech.

Register Now