Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Driving Style Alignment for LLM-powered Driver Agent

Ruoxuan Yang, Xinyue Zhang, Anais Fernandez-Laaksonen, Xin Ding, Jiangtao Gong

TL;DR

The paper addresses the challenge of aligning LL-powered driver agents with human driving styles by introducing a multi-alignment framework that leverages human demonstrations and Coach-driven feedback. A natural-language driving-behavior dataset is collected from real driving and post-drive interviews to train and evaluate alignments. The authors validate their approach in CARLA with a $3\times3$ design, showing that multi-alignment, particularly with cautious driving demonstrations, yields safer and more human-like behavior, and they corroborate these findings with extensive human evaluations. The work advances human-centric autonomous driving by enabling diverse, interpretable driving styles and highlighting the role of high-quality linguistic demonstrations in aligning embodied agents with human norms.

Abstract

Recently, LLM-powered driver agents have demonstrated considerable potential in the field of autonomous driving, showcasing human-like reasoning and decision-making abilities.However, current research on aligning driver agent behaviors with human driving styles remains limited, partly due to the scarcity of high-quality natural language data from human driving behaviors.To address this research gap, we propose a multi-alignment framework designed to align driver agents with human driving styles through demonstrations and feedback. Notably, we construct a natural language dataset of human driver behaviors through naturalistic driving experiments and post-driving interviews, offering high-quality human demonstrations for LLM alignment. The framework's effectiveness is validated through simulation experiments in the CARLA urban traffic simulator and further corroborated by human evaluations. Our research offers valuable insights into designing driving agents with diverse driving styles.The implementation of the framework and details of the dataset can be found at the link.

Driving Style Alignment for LLM-powered Driver Agent

TL;DR

The paper addresses the challenge of aligning LL-powered driver agents with human driving styles by introducing a multi-alignment framework that leverages human demonstrations and Coach-driven feedback. A natural-language driving-behavior dataset is collected from real driving and post-drive interviews to train and evaluate alignments. The authors validate their approach in CARLA with a design, showing that multi-alignment, particularly with cautious driving demonstrations, yields safer and more human-like behavior, and they corroborate these findings with extensive human evaluations. The work advances human-centric autonomous driving by enabling diverse, interpretable driving styles and highlighting the role of high-quality linguistic demonstrations in aligning embodied agents with human norms.

Abstract

Recently, LLM-powered driver agents have demonstrated considerable potential in the field of autonomous driving, showcasing human-like reasoning and decision-making abilities.However, current research on aligning driver agent behaviors with human driving styles remains limited, partly due to the scarcity of high-quality natural language data from human driving behaviors.To address this research gap, we propose a multi-alignment framework designed to align driver agents with human driving styles through demonstrations and feedback. Notably, we construct a natural language dataset of human driver behaviors through naturalistic driving experiments and post-driving interviews, offering high-quality human demonstrations for LLM alignment. The framework's effectiveness is validated through simulation experiments in the CARLA urban traffic simulator and further corroborated by human evaluations. Our research offers valuable insights into designing driving agents with diverse driving styles.The implementation of the framework and details of the dataset can be found at the link.
Paper Structure (16 sections, 3 figures)

This paper contains 16 sections, 3 figures.

Table of Contents

  1. INTRODUCTION
  2. Multi-alignment Framework
  3. Driver Agent
  4. Multi-alignment through Demonstrations and Feedback
  5. Driving Style Data Collection
  6. Natural Driving Experiment and Post-driving Interview
  7. Driving Style Selection and Data Organization
  8. Experiment
  9. Conditions
  10. CARLA Simulation
  11. Set-up
  12. Metrics
  13. Results
  14. Findings
  15. Human Evaluation
  16. ...and 1 more sections

Figures (3)

  • Figure 1: The multi-alignment framework
  • Figure 2: Experiment conditions
  • Figure 3: Simulation experiment results for predefined metrics.

Theorems & Definitions (9)

  • proof
  • proof
  • proof
  • proof
  • proof
  • proof
  • proof
  • proof
  • proof