Diffusion Models for Intelligent Transportation Systems: A Survey

TL;DR

This survey identifies diffusion models as a versatile framework for Intelligent Transportation Systems (ITS) due to their high-fidelity, controllable, and multi-modal data generation capabilities. It systematically reviews theoretical underpinnings (DDPMs, NCSNs, SDEs) and key variants (conditional and latent diffusion models), linking them to ITS tasks across autonomous driving, traffic simulation, forecasting, and safety. The paper highlights practical advantages such as robust handling of noisy data, privacy-preserving synthetic data generation, and seamless integration with graph neural networks and reinforcement learning. It also outlines current challenges in ITS and proposes several promising directions, including integrating large language models, leveraging prior knowledge for guidance, architectural innovations, fine-tuning strategies, and speedups for real-time deployment. Overall, the work aims to bridge diffusion-model advances with transportation research to enable more realistic simulations, safer planning, and scalable ITS solutions.

Abstract

Intelligent Transportation Systems (ITS) are vital in modern traffic management and optimization, significantly enhancing traffic efficiency and safety. Recently, diffusion models have emerged as transformative tools for addressing complex challenges within ITS. In this paper, we present a comprehensive survey of diffusion models for ITS, covering both theoretical and practical aspects. First, we introduce the theoretical foundations of diffusion models and their key variants, including conditional diffusion models and latent diffusion models, highlighting their suitability for modeling complex, multi-modal traffic data and enabling controllable generation. Second, we outline the primary challenges in ITS and the corresponding advantages of diffusion models, providing readers with a deeper understanding of the intersection between ITS and diffusion models. Third, we offer a multi-perspective investigation of current applications of diffusion models in ITS domains, including autonomous driving, traffic simulation, trajectory prediction, and traffic safety. Finally, we discuss state-of-the-art diffusion model techniques and highlight key ITS research directions that warrant further investigation. Through this structured overview, we aim to provide researchers with a comprehensive understanding of diffusion models for ITS, thereby advancing their future applications in the transportation domain.

Figures (7)

• Figure 1: Overview of applying diffusion models to traffic tasks using various traffic data types, including trajectories, traffic images, spatial-temporal graphs, and traffic-related texts.
• Figure 2: Different condition mechanisms for diffusion models. (1) Concatenation-based mechanism directly incorporates conditions such as historical data and maps into the input. (2) Cross-attention-based mechanism integrates conditions like text and external features through cross-attention layers. (3) Classifier-based mechanism uses an external classifier to guide denoising based on conditions such as reinforcement learning or cost functions. (4) Classifier-free mechanism combines conditional and unconditional denoising models, balancing both with a weight parameter.
• Figure 3: Illustration of latent diffusion models. Compared to standard diffusion models, they incorporate a pre-trained encoder $\mathcal{E}$ and decoder $\mathcal{D}$, with the diffusion and denoising processes operating in latent space rather than pixel or data space.
• Figure 4: The challenges in intelligent transportation systems.
• Figure 5: The advantages of diffusion models.