Current Development, Comparison and Future Directions in Vehicle Trajectory Prediction

Lian Zhou, Xiaoliang Wang, Xiaolan Zhou*, Yuzhen Liu and Hao Yue
School of Computer Science and Engineering, Hunan University of Science and Technology, China, Hunan Key Laboratory for Service Computing and Novel Software Technology, China
*Corresponding Author: Xiaolan Zhou, Hunan University of Science and Technology, Taoyuan Road, Yuhu District, Xiangtan, Hunan Province, China.

Published: April 09, 2024

View Pdf

Abstract  

With the flourishing development of autonomous driving technology and the increasing demand for convenient travel experiences, more and more researchers are diving into the development of efficient and reliable autonomous driving technology. Therefore, this paper aims to explore the main technologies of vehicle trajectory prediction in autonomous driving and comprehensively review the research status of vehicle trajectory prediction in autonomous driving over the past decade. It delves into the characteristics and differences of research based on physical methods, basic machine learning methods, and deep learning methods. Next, we focus on analyzing the currently mainstream deep learning-based vehicle trajectory prediction models. We utilize open-source datasets in the autonomous driving domain, such as the Argoverse dataset and the NuScenes dataset, as well as evaluation metrics like Average Displacement Error (ADE) and Final Displacement Error (FDE), to provide a detailed exposition and analysis of the progress made with existing technologies through research and experimentation. Finally, the article points out the possible directions for future breakthroughs in this field, aiming to guide readers and researchers to overcome existing technological bottlenecks and further promote the advancement of this field.

Keywords: Autonomous Driving; Vehicle Trajectory Prediction; Physics-based Methods; Machine Learning; Deep Learning

.