EquiFusion: Kinematics-Agnostic Human Motion Prediction via Equivariant Latent Diffusion

2026-07-12

AI Summary

This work addresses the limitations of existing stochastic 3D human motion prediction methods, which are hindered by hard-coded skeletal structures that impede generalization, prevent cross-dataset training, and rely on complex retargeting. The authors propose the first kinematics-agnostic prediction framework, explicitly conditioning on skeletal connectivity and integrating it into a permutation-equivariant latent diffusion architecture. This design inherently achieves invariance to joint ordering and graph topology, enabling zero-shot cross-dataset generalization. The model effectively handles partial observations, occlusions, and task-specific limb generation, achieving state-of-the-art performance on standard benchmarks while reducing model size by up to 75%, thereby significantly improving both training and inference efficiency.

Abstract

Existing Stochastic 3D Human Motion Prediction models are fundamentally constrained by hard-coding the skeleton kinematics, severely limiting generalization, preventing cross-dataset training, and requiring complex data retargeting. We introduce EquiFusion, the first kinematics-agnostic model to solve this bottleneck, implementing a latent diffusion model with a permutation equivariant architecture. EquiFusion treats the kinematics' connectivity as an explicit input parameter, ensuring its internal computations are inherently agnostic to joint ordering and graph structure. This novel design enables truly cross-dataset generalization to unseen kinematics and unlocks novel zero-shot directions, such as motion prediction from partial or occluded observations and targeted limb generation. EquiFusion achieves state-of-the-art results on major benchmarks, being up to 75% more compact than previous kinematics-specific methods, while achieving faster training and inference. EquiFusion thus establishes a new, flexible standard for robust human motion prediction. Model and training code are available at https://ceveloper.github.io/publications/equifusion/.

Problem

• Human Motion Prediction
• Kinematics-Agnostic
• Cross-Dataset Generalization
• Skeleton Kinematics
• Stochastic 3D Modeling

Innovation

• kinematics-agnostic
• equivariant diffusion
• latent diffusion model
• cross-dataset generalization
• human motion prediction