Artificial Intelligence for Modeling and Simulation of Mixed Automated and Human Traffic

A comprehensive survey examines AI methodologies for simulating mixed autonomous and human-driven traffic, addressing critical gaps in current simulation tools. The research proposes a unified taxonomy of AI methods spanning agent-level behavior models, environment-level simulations, and physics-informed approaches to improve autonomous vehicle testing and validation.

This survey addresses a fundamental challenge in autonomous vehicle development: the need for realistic, AI-powered traffic simulation that accurately captures both AV and human driver behaviors. Current simulation platforms prioritize graphical fidelity while relying on oversimplified rule-based models, failing to represent the nuanced complexity of real-world driving interactions. As AVs move from controlled testing to public roads, the inability to accurately simulate mixed autonomy scenarios creates significant validation gaps for safety-critical systems.

The research synthesizes AI methods across three architectural layers—agent-level behavior, environment-level dynamics, and cognitive physics modeling—providing a structured framework that bridges traffic engineering and computer science perspectives. This interdisciplinary approach recognizes that autonomous vehicle safety depends on understanding not just individual decision-making, but emergent traffic patterns arising from human-AV interaction. The survey's taxonomy enables researchers to identify which AI methodologies (reinforcement learning, transformer networks, physics-informed neural networks) best suit specific simulation requirements.

For the AV industry, improved simulation capabilities accelerate development cycles by enabling safer, more comprehensive testing before real-world deployment. Simulation platforms that integrate advanced AI behavior modeling reduce the computational burden of edge-case testing and provide statistical confidence in safety metrics. Investment in simulation technology directly correlates with AV commercialization timelines and insurance/regulatory approval processes.

Looking forward, the critical metric becomes whether simulation environments can achieve sufficient fidelity to replace extensive real-world testing. Adoption of unified evaluation protocols and standardized datasets becomes essential for cross-platform validation. The field's maturation depends on closing the gap between academic AI research and production simulation platforms.

• →Existing traffic simulation tools lack AI-based behavior modeling, relying on simplistic rules that fail to capture real driving complexity.
• →The survey introduces a three-tier taxonomy organizing AI methods from individual agent behavior to full-scene simulation.
• →Mixed autonomy simulation remains underexplored despite AVs now operating on public roads, creating validation gaps.
• →Bridging traffic engineering and computer science perspectives is essential for developing realistic autonomous vehicle testing environments.
• →Advanced AI simulation reduces real-world testing requirements and accelerates autonomous vehicle commercialization timelines.