IDM-Follower: A Model-Informed Deep Learning Method for Car-Following Trajectory Prediction
Yilin Wang, Yiheng Feng
Abstract
Model-based and learning-based methods are two main approaches modeling car-following behaviors. To combine advantages from both types of models, this study introduces a novel approach, IDM-Follower, which generates a sequence of the following vehicle's trajectory using a recurrent autoencoder informed by a physical car-following model, the Intelligent Driving Model (IDM). We design an innovative neural network (NN) structure with two independent encoders and an attentionbased decoder to predict the trajectory sequence. The loss function accounts for discrepancies from both the physical carfollowing model and the NN predictions. Numerical experiments are conducted using simulated and real world (i.e., NGSIM) datasets under different data noise levels with varying weights between the learning loss and the model loss. The prediction results show the proposed approach outperforms both modelbased and learning-based baselines under real and high noise levels. The optimal weight between the model and learning component is strongly related to the data quality and model accuracy.