Work

Grounding the Phantom: Improving Physical Plausibility in Dance Generation via Biomechanical Constraints

Year: 2026

Type: conference-paper

Abstract: Music-conditioned 3D dance generation often suffers from physical artifacts such as foot sliding. We propose a Physics-Informed Loss Module that integrates constraints on static friction (PFC-Proxy), trajectory smoothness (Minimum Jerk), and dynamic balance (ZMP) into a diffusion model. Experiments on AIST++ demonstrate that our approach reduces foot skating by 45% and achieves physical plausibility scores that surpass those of the ground-truth data, significantly enhancing the realism of the generated dances.

Source: IEEE International Conference on Consumer Electronics - Taiwan (ICCE-Taiwan)

DOI: 10.1109/icce-taiwan71481.2026.11652385

Authors: Hsing-Hao Yeh, Feng-Cheng Lin, Cheng-Tai Jiang

Institution: Feng Chia University

Language: English

Topic: Human Motion and Animation

Subfield: Computer Graphics and Computer-Aided Design

Field: Computer Science

Domain: Physical Sciences

Open Access status: hybrid

OpenAlex ID: W7204120534

FWCI:

Cites: 6

Cited by:

Related to: 10