UniffordInteractor is a two-stage framework for generating human motion when a person interacts with several objects, and the authors say it beats existing methods.
Reasoning stage: A Unified Affordance Reasoner uses a fine-tuned 3D Multimodal Large Language Model to predict object-centric affordances, including object-object contact and not just human-object contact.
Interaction stage: A Hierarchical Interaction Generator builds motion autoregressively from root to body to hands to objects, conditioned on those affordances.
Claimed results: The paper says it surpasses state-of-the-art methods on human and multi-object interaction generation, with long-horizon scenes as the focus.
Paper details: arXiv v1, submitted 29 October 2026, 13 pages and 11 figures, listed as a SIGGRAPH Asia 2026 paper in ACM Transactions on Graphics.
Sign in to comment.
TL;DR of the paper, generated automatically.
UniffordInteractor is a two-stage framework for generating human motion when a person interacts with several objects, and the authors say it beats existing methods.