Generative AI in Transportation Planning: A Survey

Longchao Da, Tiejin Chen, Zhuoheng Li, Shreyas Bachiraju, Huaiyuan Yao, Yushun Dong, Xiyang Hu, Zhengzhong Tu, Dongjie Wang, Yue Zhao, Xuanyu Zhou, Ram Pendyala, Benjamin Stabler, Yezhou Yang, Xuesong Zhou, Hua Wei

Abstract

The integration of generative artificial intelligence (GenAI) into transportation planning has the potential to revolutionize tasks such as demand forecasting, infrastructure design, policy evaluation, and traffic simulation. However, there is a critical need for a systematic framework to guide the adoption of GenAI in this interdisciplinary domain. In this survey, we present the first comprehensive framework for leveraging GenAI in transportation planning. We introduce a new taxonomy that categorizes existing applications and methodologies into two perspectives: transportation planning tasks and computational techniques. From the planning perspective, we examine GenAI's role in descriptive, predictive, generative, and simulation tasks. From the computational perspective, we detail advancements in data preparation, domain-specific fine-tuning, and inference strategies. Additionally, we address critical challenges, including data scarcity, explainability, and bias mitigation, aiming to bridge the gap between traditional methodologies and modern AI techniques.

Introduction

Transportation planning is a data-intensive discipline that traditionally relies on expert-driven, static statistical models. These methods often struggle to scale, fail to incorporate heterogeneous data sources—such as social media or real-time sensor feeds—and lack the flexibility to simulate complex "what-if" scenarios rapidly. While Large Language Models (LLMs) and other GenAI tools offer transformative potential, the field faces significant gaps: a lack of systematic frameworks, domain-specific challenges like model "hallucinations," and the difficulty of integrating complex regulatory knowledge into general-purpose models.

This paper addresses these gaps by proposing a dual-perspective taxonomy that organizes the "GenAI for Transportation" landscape, bridging traditional engineering tasks with modern computational techniques.

Organization of the survey framework and its thematic sections

A Dual-Perspective Taxonomy

The Planning Perspective

The planning perspective categorizes tasks based on their functional role in the transportation ecosystem. GenAI helps enhance these domains through automation, scalability, and adaptability:

  • Descriptive Tasks: Data fusion and analysis, including traffic monitoring and estimation.
  • Predictive Tasks: Travel demand management (OD calibration) and traffic prediction (human and vehicle-related).
  • Generative & Simulation Tasks: Synthesis of traffic scenarios, driving behavior simulation, and environmental modeling (emission forecasting).
  • Decision-Making & Optimization: Route planning, traffic signal control, and freight logistics optimization.
The benefits of GenAI across various transportation planning domains

The Computational Perspective

This perspective details the technical methodologies required to adapt GenAI for transportation:

  • Fine-Tuning: Utilizing Parameter-Efficient Fine-Tuning (PEFT) techniques like LoRA to adapt models to specific datasets, such as Origin-Destination (OD) demand matrices.
  • Inference Strategies: Leveraging Zero-Shot Learning (ZSL) for accessibility metrics and Few-Shot Learning (FSL) for multi-agent traffic coordination.
  • Advanced Architectures: Implementing Retrieval-Augmented Generation (RAG) to ground model outputs in real-time traffic reports and Chain-of-Thought (CoT) reasoning for multi-step optimization.
  • ExpertLLM: A novel approach that uses domain heuristics to decompose complex engineering problems into manageable tasks for the LLM.

Key Insights and Results

Domain Knowledge is Essential

The authors demonstrate that "Vanilla LLMs" (generic models) often fail at complex numerical tasks like OD calibration, sometimes producing unusable "all-zero" matrices. Success requires an ExpertLLM pipeline that uses engineering logic to guide the AI. Notably, a fine-tuned 8B parameter model (e.g., Llama-3) can outperform a generic 70B parameter model on domain-specific tasks, proving that data quality and task-specific tuning are more critical than raw model size.

Performance Benchmarks

  • OD Calibration: The ExpertLLM architecture achieved significant reductions in Mean Squared Error (MSE) compared to traditional Genetic Algorithms and generic LLM approaches.
  • Traffic Coordination: Using the CoMAL framework (Few-Shot Learning), AI-guided autonomous vehicles demonstrated higher average speeds and lower speed variance than human drivers in complex merging scenarios.
  • Accessibility Estimation: In zero-shot tests across 85 global cities, LLMs estimated "People Near Transit" (PNT) scores with high correlation to ground-truth GIS data, suggesting they can serve as low-cost proxies for data-scarce regions.

Challenges and Future Directions

Despite the potential, several "pitfalls" remain. Data scarcity for high-quality, domain-specific open datasets leads to model bias. Furthermore, models can still suggest "plausible but impossible" infrastructure designs (hallucinations).

To ensure adoption by public agencies, the authors propose a "Trustworthiness Hexagon" consisting of Privacy, Security, Fairness, Responsibility, Explainability, and Reliability. Future research must focus on Agentic RAG pipelines—where AI agents autonomously retrieve data, run simulations, and refine policy recommendations—and the development of novel evaluation criteria that align with goals like sustainability and equity.

Conclusion

GenAI represents a paradigm shift for transportation planning, offering the tools to democratize expertise and foster more equitable public engagement. By bridging the gap between AI research and transportation engineering, this survey provides a roadmap for the ethical and impactful use of generative models in building the mobility systems of the future.

Summary updated

Abstract

The integration of generative artificial intelligence (GenAI) into transportation planning has the potential to revolutionize tasks such as demand forecasting, infrastructure design, policy evaluation, and traffic simulation. However, there is a critical need for a systematic framework to guide the adoption of GenAI in this interdisciplinary domain. In this survey, we, a multidisciplinary team of researchers spanning computer science and transportation engineering, present the first comprehensive framework for leveraging GenAI in transportation planning. Specifically, we introduce a new taxonomy that categorizes existing applications and methodologies into two perspectives: transportation planning tasks and computational techniques. From the transportation planning perspective, we examine the role of GenAI in automating descriptive, predictive, generative, simulation, and explainable tasks to enhance mobility systems. From the computational perspective, we detail advancements in data preparation, domain-specific fine-tuning, and inference strategies, such as retrieval-augmented generation and zero-shot learning tailored to transportation applications. Additionally, we address critical challenges, including data scarcity, explainability, bias mitigation, and the development of domain-specific evaluation frameworks that align with transportation goals like sustainability, equity, and system efficiency. This survey aims to bridge the gap between traditional transportation planning methodologies and modern AI techniques, fostering collaboration and innovation. By addressing these challenges and opportunities, we seek to inspire future research that ensures ethical, equitable, and impactful use of generative AI in transportation planning.
BibTeX
@misc{da2025generative,
  author = {Longchao Da and Tiejin Chen and Zhuoheng Li and Shreyas Bachiraju and Huaiyuan Yao and Li Li and Yushun Dong and Xiyang Hu and Zhengzhong Tu and Dongjie Wang and Yue Zhao and Xuanyu Zhou and Ram Pendyala and Benjamin Stabler and Yezhou Yang and Xuesong Zhou and Hua Wei},
  title = {{Generative AI in Transportation Planning: A Survey}},
  year = {2025},
  eprint = {2503.07158},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2503.07158},
}
Figure