MARRS framework generates coordinated human reaction motions using continuous representations

Researchers have developed MARRS, a novel framework for synthesizing human reactions conditioned on observed actions. The system utilizes a Unit-distinguished Motion Variational AutoEncoder (UD-VAE) to encode distinct body and hand units independently. It incorporates Action-Conditioned Fusion (ACF) to process reactive tokens and Mutual Unit Modulation (MUM) to enable interaction between body and hand units. A compact MLP serves as a noise predictor within a diffusion model for generating token probability distributions. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Introduces a new method for generating coordinated human reaction motions, potentially improving embodied AI and animation.

RANK_REASON This is a research paper detailing a novel framework for human action-reaction synthesis.

Read on arXiv cs.CV →

• MARRS
• UD-VAE
• ACF
• MUM
• arXiv

• paper
• other