Buyer, Johannes (Heilbronn University of Applied Sciences), Wetzel, Konstantin (Hochschule Heilbronn), Zöllner, Raoul (Universtiy of Heilbronn), Zöllner, J. Marius (FZI Research Center for Information Technology; KIT Karlsruhe In)

Longitudinal Vehicle Motion Prediction in Traffic Junctions using a Data-Driven Multi-object Intelligent Driver Model

Scheduled for presentation during the Regular Session "Traffic prediction and estimation I" (WeAT6), Wednesday, September 25, 2024, 12:10−12:30, Salon 14

2024 IEEE 27th International Conference on Intelligent Transportation Systems (ITSC), September 24- 27, 2024, Edmonton, Canada

This information is tentative and subject to change. Compiled on October 8, 2024

Abstract

The paper presents an approach for longitudinal interaction-aware vehicle motion prediction using an Intelligent Driver Model (IDM) with a data-driven extension, which is able to consider several traffic participants. The basic prediction concept for that approach was originally designed for multi-lane traffic scenarios in the authors' previous work. The approach is based on model knowledge, probabilistic estimation as well as machine learning components. The innovation of the paper at hand is the transfer of the concept for its usage in traffic junctions. Thereby, a decision tree is developed that processes different interaction features in order to select the relevant traffic participants. To estimate the model parameter values, a Sequential Importance Resampling (SIR) particle filter is used. Time-varying parameters which weight the selected objects considered in the multi-object IDM are adapted over the prediction horizon using a trained support vector machine (SVM) regression model. For validation, roundabout motion trajectories of the INTERACTION dataset are used. Since the proposed prediction approach is able to consider vehicles' turn signal status, an augmentation of the dataset with turn signal attributes is realized. The accuracy of the prediction approach is compared to baseline methods by calculating the mean absolute error (MAE) of predicted vehicle speeds.