FunPhase: A Periodic Functional Autoencoder for Motion Generation via Phase Manifolds

2025-12-10

AI Summary

Modeling natural human motion faces the challenge of strong coupling between spatial geometry and temporal dynamics. Existing phase-manifold-based approaches capture local periodicity but suffer from poor generalizability, sensitivity to dataset-specific configurations, and inability to support arbitrary temporal resolutions. This paper introduces the first functional-space periodic autoencoder, which lifts discrete-time decoding to a continuous function mapping—thereby constructing a generalizable phase manifold—and unifies motion prediction and generation within a single framework enabling cross-skeleton and cross-dataset transfer. Our method integrates functional neural networks, periodic latent-space learning, and a functional decoder to achieve smooth motion reconstruction, temporal super-resolution, and occluded limb completion. Experiments demonstrate significantly lower reconstruction error than baseline periodic autoencoders and state-of-the-art generation quality.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Computer Vision: Motion & Tracking
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

Learning natural body motion remains challenging due to the strong coupling between spatial geometry and temporal dynamics. Embedding motion in phase manifolds, latent spaces that capture local periodicity, has proven effective for motion prediction; however, existing approaches lack scalability and remain confined to specific settings. We introduce FunPhase, a functional periodic autoencoder that learns a phase manifold for motion and replaces discrete temporal decoding with a function-space formulation, enabling smooth trajectories that can be sampled at arbitrary temporal resolutions. FunPhase supports downstream tasks such as super-resolution and partial-body motion completion, generalizes across skeletons and datasets, and unifies motion prediction and generation within a single interpretable manifold. Our model achieves substantially lower reconstruction error than prior periodic autoencoder baselines while enabling a broader range of applications and performing on par with state-of-the-art motion generation methods.

Problem

• Learning natural body motion with spatial-temporal coupling
• Embedding motion in scalable phase manifolds for prediction
• Enabling smooth motion generation at arbitrary resolutions

Innovation

• Functional autoencoder learns phase manifold for motion
• Replaces discrete decoding with function-space formulation
• Unifies prediction and generation in interpretable manifold