LLM-based Human-like Traffic Simulation for Self-driving Tests

TL;DR

This paper presents HDSim, a novel traffic simulation framework combining cognitive theory and large language models to generate diverse, realistic traffic scenarios for testing self-driving cars, improving safety failure detection.

Contribution

HDSim introduces a hierarchical driver model and perception-guided LLM strategies, advancing traffic simulation realism and interpretability over existing methods.

Findings

Embedding HDSim improves safety failure detection by up to 68%.

HDSim produces more realistic and interpretable traffic scenarios.

The framework enhances diversity and realism in traffic simulation.

Abstract

Ensuring realistic traffic dynamics is a prerequisite for simulation platforms to evaluate the reliability of self-driving systems before deployment in the real world. Because most road users are human drivers, reproducing their diverse behaviors within simulators is vital. Existing solutions, however, typically rely on either handcrafted heuristics or narrow data-driven models, which capture only fragments of real driving behaviors and offer limited driving style diversity and interpretability. To address this gap, we introduce HDSim, an HD traffic generation framework that combines cognitive theory with large language model (LLM) assistance to produce scalable and realistic traffic scenarios within simulation platforms. The framework advances the state of the art in two ways: (i) it introduces a hierarchical driver model that represents diverse driving style traits, and (ii) it develops a Perception-Mediated Behavior Influence strategy, where LLMs guide perception to indirectly shape driver actions. Experiments reveal that embedding HDSim into simulation improves detection of safety-critical failures in self-driving systems by up to 68% and yields realism-consistent accident interpretability.