Discrete Contrastive Learning for Diffusion Policies in Autonomous Driving

2025-03-07

AI Summary

To address the challenge of modeling human driving behavior diversity in autonomous driving simulation, this paper proposes a discretized contrastive learning framework. First, it integrates contrastive learning with vector quantization (VQ) to learn an interpretable, quantifiable driving style dictionary directly from real-world driving data. Second, it constructs a style-conditioned diffusion policy model to generate high-fidelity, safe, and human-like driving behaviors. The method unifies behavioral cloning, VQ-based representation learning, and conditional diffusion modeling, enabling controllable and interpretable behavior synthesis. Experiments demonstrate that the generated behaviors reduce collision rates by 23% and improve expert-rated human-likeness scores by 31% in simulation, significantly outperforming mainstream machine learning baselines. This advancement enhances both the realism of autonomous driving testing and the validity of evaluation metrics.

Technology Category

• Humans and AI: Human-Aware Planning and Behavior Prediction
• Cognitive Modeling & Cognitive Systems: Simulating Human Behavior
• Computer Vision: Diffusion Models for Vision

Application Category

• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Learning to perform accurate and rich simulations of human driving behaviors from data for autonomous vehicle testing remains challenging due to human driving styles' high diversity and variance. We address this challenge by proposing a novel approach that leverages contrastive learning to extract a dictionary of driving styles from pre-existing human driving data. We discretize these styles with quantization, and the styles are used to learn a conditional diffusion policy for simulating human drivers. Our empirical evaluation confirms that the behaviors generated by our approach are both safer and more human-like than those of the machine-learning-based baseline methods. We believe this has the potential to enable higher realism and more effective techniques for evaluating and improving the performance of autonomous vehicles.

Problem

• Simulate diverse human driving behaviors accurately
• Extract and discretize driving styles using contrastive learning
• Improve safety and realism in autonomous vehicle testing

Innovation

• Contrastive learning extracts driving styles
• Quantization discretizes diverse driving behaviors
• Conditional diffusion policy simulates human drivers