Pedestrian Behavior Modeling

The field studies pedestrian movement, crossing intentions, social interactions, crowd flow, and responses to vehicles, robots, and hazards. Common approaches include social-force and agent-based models, graph and Transformer networks, state-space models, and probabilistic methods such as conditional variational autoencoders and diffusion models; language models are also used for scenario generation and decision-making. Methods are evaluated on trajectory datasets such as ETH/UCY, SDD, JRDB, JAAD, and PIE, as well as in simulations, robot deployments, and evacuation settings. A central challenge is making predictions both accurate and behaviorally realistic across unfamiliar environments, while representing uncertainty and evaluating more than displacement error alone.