Interact2Ar: Full-Body Human-Human Interaction Generation via Autoregressive Diffusion Models
Pablo Ruiz-Ponce, Sergio Escalera, José García-Rodríguez, Jiankang Deng, Rolandos Alexandros Potamias · 2025
Read Interact2Ar: Full-Body Human-Human Interaction Generation via Autoregressive Diffusion Models by Pablo Ruiz-Ponce, Sergio Escalera, José García-Rodríguez, Jiankang Deng, Rolandos Alexandros Potamias (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2512-19692.txt.
Description
Pablo Ruiz-Ponce and four colleagues built Interact2Ar, a generator for two people moving together, hands included. Earlier motion generators often skipped hands entirely and produced whole sequences at once, which flattens the reactive, back-and-forth quality of real interaction. Data limits and learning difficulty pushed prior work toward those shortcuts.
Interact2Ar is text-conditioned and autoregressive, generating motion step by step with a diffusion model, while dedicated parallel branches handle detailed hand kinematics. A memory technique with large context windows lets it adapt as interactions unfold. The design supports follow-on uses like composing motion over time, reacting to disturbances in real time, and scaling beyond pairs to groups. New evaluators and metrics for full-body interaction back claims of state-of-the-art quality.
Catalog record
| CREATOR | Pablo Ruiz-Ponce, Sergio Escalera, José García-Rodríguez, Jiankang Deng, Rolandos Alexandros Potamias |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 59 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2512-19692