Discovering Car-following Dynamics from Trajectory Data through Deep Learning
Ohay Angah, James Enouen, Xuegang, Ban, Yan Liu
TL;DR
The paper addresses the challenge of discovering governing car-following dynamics from trajectory data in a way that yields interpretable, concise expressions. It introduces a VIS-enhanced deep symbolic regression framework that first identifies informative variable interactions via VIS and then searches for governing expressions with a risk-seeking LSTM, guided by a reward that penalizes complexity and enforces VIS-consistency. The approach recovers classic Krauss/Gaussian-like dynamics under clean data and demonstrates robustness to noise, outperforming standard DSR and GP-assisted variants in efficiency and interpretability. The work offers a path toward automated, interpretable discovery of traffic dynamics and lays groundwork for applying similar methods to multimodal and evolving traffic environments.
Abstract
This study aims to discover the governing mathematical expressions of car-following dynamics from trajectory data directly using deep learning techniques. We propose an expression exploration framework based on deep symbolic regression (DSR) integrated with a variable intersection selection (VIS) method to find variable combinations that encourage interpretable and parsimonious mathematical expressions. In the exploration learning process, two penalty terms are added to improve the reward function: (i) a complexity penalty to regulate the complexity of the explored expressions to be parsimonious, and (ii) a variable interaction penalty to encourage the expression exploration to focus on variable combinations that can best describe the data. We show the performance of the proposed method to learn several car-following dynamics models and discuss its limitations and future research directions.