Techniques to Enhance Learning-based Trajectory Prediction
Handong Yao, Ph.D.
Friday, January 21, 2022
12:00 PM – 1:00 PM EST
The physics of shockwaves is a fundamental traffic characteristic that is useful for microscopic traffic flow modeling. Numerous classical physics-based models have utilized the physics of shockwaves to predict vehicle trajectory dynamics, yet their predictability is often limited due to the volatile and complex nature of highway traffic composed of human-driven vehicles. Recent learning-based trajectory prediction models utilize historical trajectories of surrounding vehicles to improve predictability. However, those learning-based models are purely data-driven, thus lacking interpretability and physical insights, or even missing opportunities for further improving model predictability. To leverage the advantages of both learning-based and physics-based models, this study proposes a physics-aware learning-based model for a trajectory prediction of congested traffic in a connected vehicle environment. A newly collected highway trajectory dataset is adopted for training and validation. Experiment results show that the proposed hybrid model yields better predictability, compared with the learning-based models (e.g., long short-term memory neural networks and convolutional neural networks), with an 8.7% predictability reduction of position errors, which further verify the positive impacts of adopting physics of shockwaves in hybrid learning models.
Handong Yao, Ph.D., is a postdoctoral research associate in the Department of Civil and Environmental Engineering at the University of South Florida. He received his Ph.D. and M.S. degrees in transportation engineering from Harbin Institute of Technology, Harbin, China. Dr. Yao has 9 years of experience in connected & automated vehicles (CAV) and traffic flow modeling. Dr. Yao has cooperated with Dr. Xiaopeng Li for more than five years at the Connected & Autonomous Transportation Systems (CATS) lab. Dr. Yao has worked as a Co-PI in several projects supported by multiple agencies (e.g., NSF, US DOT, US DOE). His key research interests include machine-learning-based data analysis and traffic operations with CAVs.