A Physics Enhanced Residual Learning (PERL) Framework for Vehicle Trajectory Prediction

Monday, June 17, 2024

1:30 PM - 2:45 PM ET

Room: Galleria

Primary Author(s)

Keke Long, MA

Research Assistant
University of Wisconsin–Madison
Madison, WI, United States

Presenter(s)

Keke Long, MA

Research Assistant
University of Wisconsin–Madison
Madison, WI, United States

Co-Author(s)

HS

Haotian Shi, PhD (he/him/his)

Research Associate
UW-Madison
Madison, WI, United States
ZS

Zihao Sheng, Master

Research Assistant
University of Wisconsin, Madison
MADISON, Wisconsin, United States
Xiaopeng Li, PhD

Professor
University of Wisconsin–Madison, United States
Sikai Chen

Assistant Professor
Department of Civil and Environmental Engineering, University of Wisconsin-Madison, United States

Submitter(s)

Keke Long, MA

Research Assistant
University of Wisconsin–Madison
Madison, WI, United States

Abstract:

In vehicle trajectory prediction, physics models and data-driven models are two predominant methodologies. However, each approach presents its own set of challenges: physics models fall short in predictability, while data-driven models lack interpretability. Addressing these identified shortcomings, this paper proposes a novel framework, the Physics-Enhanced Residual Learning (PERL) model. PERL integrates the strengths of physics-based and data-driven methods for vehicle trajectory prediction. PERL contains a physics model and a residual learning model. Its prediction is the sum of the physics model result and a predicted residual as a correction to it. It preserves the interpretability inherent to physics-based models and has reduced data requirements compared to data-driven methods. Experiments were conducted using a real-world vehicle trajectory dataset. We proposed a PERL model, with the Intelligent Driver Model (IDM) as its physics car-following model and Long Short-Term Memory (LSTM) as its residual learning model. We compare this PERL model with the physics car-following model, data-driven model, and other physics-informed neural network (PINN) models. The result reveals that PERL achieves better prediction with a small dataset, compared to the physics model, data-driven model, and PINN model. Second, the PERL model showed faster convergence during training, offering comparable performance with fewer training samples than the data-driven model and PINN model. Sensitivity analysis also proves comparable performance of PERL using another residual learning model and a physics car-following model.

Learning Objectives:

Attendees can expect to learn the following from this session:

Upon completion, participants will be able to know the core principles and advantages of the Physics-Enhanced Residual Learning (PERL) framework over traditional PINN and data-driven models.
Upon completion, participants will be able to describe the mechanism and significance of residual learning in trajectory prediction, emphasizing its utility in enhancing model predictability and efficiency.
Upon completion, participants will be able to know how to apply PERL to real-world trajectory datasets to assess the performance and superiority of the PERL model.