Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GenFollower: Enhancing Car-Following Prediction with Large Language Models

Xianda Chen, Mingxing Peng, PakHin Tiu, Yuanfei Wu, Junjie Chen, Meixin Zhu, Xinhu Zheng

TL;DR

GenFollower reframes car-following prediction as a language-modeling problem and uses zero-shot prompting of large language models to predict following-vehicle speed while generating explicit explanations. Evaluated on the Waymo Open Dataset, it achieves competitive accuracy and a strong safety profile, including 0% collisions for GPT-4 in the prompt-based setting, and provides interpretable outputs through step-by-step reasoning. The work demonstrates the potential of LLM-based prompting for interpretable, data-efficient autonomous-driving components and discusses the trade-offs with fine-tuning. Overall, GenFollower represents a pioneering approach that combines zero-shot prompting with interpretability to advance car-following prediction for safer and more transparent autonomous systems.

Abstract

Accurate modeling of car-following behaviors is essential for various applications in traffic management and autonomous driving systems. However, current approaches often suffer from limitations like high sensitivity to data quality and lack of interpretability. In this study, we propose GenFollower, a novel zero-shot prompting approach that leverages large language models (LLMs) to address these challenges. We reframe car-following behavior as a language modeling problem and integrate heterogeneous inputs into structured prompts for LLMs. This approach achieves improved prediction performance and interpretability compared to traditional baseline models. Experiments on the Waymo Open datasets demonstrate GenFollower's superior performance and ability to provide interpretable insights into factors influencing car-following behavior. This work contributes to advancing the understanding and prediction of car-following behaviors, paving the way for enhanced traffic management and autonomous driving systems.

GenFollower: Enhancing Car-Following Prediction with Large Language Models

TL;DR

GenFollower reframes car-following prediction as a language-modeling problem and uses zero-shot prompting of large language models to predict following-vehicle speed while generating explicit explanations. Evaluated on the Waymo Open Dataset, it achieves competitive accuracy and a strong safety profile, including 0% collisions for GPT-4 in the prompt-based setting, and provides interpretable outputs through step-by-step reasoning. The work demonstrates the potential of LLM-based prompting for interpretable, data-efficient autonomous-driving components and discusses the trade-offs with fine-tuning. Overall, GenFollower represents a pioneering approach that combines zero-shot prompting with interpretability to advance car-following prediction for safer and more transparent autonomous systems.

Abstract

Accurate modeling of car-following behaviors is essential for various applications in traffic management and autonomous driving systems. However, current approaches often suffer from limitations like high sensitivity to data quality and lack of interpretability. In this study, we propose GenFollower, a novel zero-shot prompting approach that leverages large language models (LLMs) to address these challenges. We reframe car-following behavior as a language modeling problem and integrate heterogeneous inputs into structured prompts for LLMs. This approach achieves improved prediction performance and interpretability compared to traditional baseline models. Experiments on the Waymo Open datasets demonstrate GenFollower's superior performance and ability to provide interpretable insights into factors influencing car-following behavior. This work contributes to advancing the understanding and prediction of car-following behaviors, paving the way for enhanced traffic management and autonomous driving systems.
Paper Structure (26 sections, 4 equations, 7 figures, 2 tables)

This paper contains 26 sections, 4 equations, 7 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Traditional Car-Following Models
  4. Data-Driven Car-Following Models
  5. LLMs for Autonomous Driving
  6. Methodology
  7. Problem Formulation
  8. GenFollower Pipeline
  9. Prompt Design
  10. System Message
  11. User Message
  12. Prompt Engineering vs. Fine-Tuning
  13. Transparency in Predictive Explanations
  14. Data and Experiments
  15. Waymo Open Dataset
  16. ...and 11 more sections

Figures (7)

  • Figure 1: Overall architecture of the GenFollower model for predicting car-following behavior in autonomous driving systems.
  • Figure 2: Visualization of car-following event and the four dimensions of the input feature.
  • Figure 3: Pipeline of the GenFollower model, which nvolves several stages to transform raw data into meaningful predictions and explanations.
  • Figure 4: System message for predicting car-following behavior in autonomous driving systems.
  • Figure 5: User message for predicting car-following behavior in autonomous driving systems.
  • ...and 2 more figures