The field of urban dynamics and mobility prediction is rapidly evolving, with a focus on developing innovative models and techniques to accurately forecast complex urban phenomena. Recent research has emphasized the importance of incorporating spatial structure, multimodal data, and uncertainty estimation into predictive models. Graph neural networks, diffusion-based models, and transformer architectures have emerged as promising approaches for modeling urban dynamics, traffic flow, and pedestrian behavior. These models have shown significant improvements in prediction performance, scalability, and interpretability. Notable papers in this area have proposed novel frameworks for predicting large-scale urban network dynamics, optimizing sensor placement for bicycling volume estimation, and generating accurate and diverse trajectory predictions for pedestrians and vehicles.

Noteworthy papers include: Predicting Large-scale Urban Network Dynamics with Energy-informed Graph Neural Diffusion, which presents a principled interpretable neural diffusion scheme for predicting urban network dynamics. TopoDiffuser: A Diffusion-Based Multimodal Trajectory Prediction Model with Topometric Maps, which introduces a diffusion-based framework for multimodal trajectory prediction that incorporates topometric maps. INSPIRE-GNN: Intelligent Sensor Placement to Improve Sparse Bicycling Network Prediction via Reinforcement Learning Boosted Graph Neural Networks, which proposes a novel RL-boosted hybrid GNN framework for optimizing sensor placement and improving link-level bicycling volume estimation.