2025 PSR UTC Annual Congress

Deep Generative Model for Human Mobility Behavior - Modeling and simulating human mobility lies at the core of numerous applications, ranging from transport planning through sustainable urban design all the way to public health. Despite ongoing efforts, accurately simulating individual mobility remains challenging, primarily due to the intricate interactions and inherent exploratory tendencies underlying human movement decisions. Here, we introduce MobilityGen, a deep generative model designed to comprehensively represent individual mobility, simulating participation in mobility events over periods ranging from days to weeks and across large spatial scales. MobilityGen captures the interactions between behavioral attributes and related contexts, and effectively reproduces key characteristics of mobility behavior, including observed scaling laws for location visits, activity times, and the evolution of travel mode and location choices. We further demonstrate that MobilityGen accounts for spatio-temporal variability in mobility behavior and generates diverse, plausible, and novel mobility events aligned with contextual information about the physical environment. By adopting deep generative modeling, our work establishes a new paradigm for mobility simulation, paving the way for novel insights into mobility behavior and supporting applications that require fine-grained mobility data.