IDM-Follower: A Physics-Informed Neural Network Model for Trajectory Prediction

Role: Individual Research Project Period: Jan 2022 – Feb 2024 Code: GitHub

Overview

How to integrate explicit car-following models into learning-based models to improve training and prediction performance.

Contributions

• Introduced a physics-informed neural network (PINN) model that integrates the Intelligent Driving Model (IDM).
• The proposed model exhibits robustness against real-time GPS noise.

Methodology

The loss function combining physical laws and ground-truth difference is utilized to train a customized attention-based VAE model.

Outcomes

• Presentation: IEEE Intelligent Vehicles Symposium (IV) 2024, Jeju Island, South Korea.
• Publication: IDM-Follower: A Model-Informed Deep Learning Method for Car-Following Trajectory Prediction, IEEE Transactions on Intelligent Vehicles, vol. 9, no. 6, pp. 5014-5020, June 2024.