Object-Conditioned Social Diffusion for Multi-Person Motion Forecasting
August 28, 2026
Object-Conditioned Social Diffusion (OCSD) integrates motion history, multi-person social encoders, and object cues into a single conditional diffusion framework. The model modulates denoising at every timestep to enable fine-grained reasoning between humans and objects in complex environments.
HOW THIS AFFECTS YOU
●
builderThis provides a method for generating plausible multi-person movement in crowded, object-rich digital environments.
●
researcherYou can use this architecture to better model human-object interactions in multi-agent simulations.