MAMM: Motion Control via Metric-Aligning Motion Matching

Control motion by
without training, annotations or large datasets

Naoki Agata1, Takeo Igarashi1

The University of Tokyo1

SIGGRAPH 2025

Video

Algorithm Overview

Our method optimizes a fused semi-unbalanced Gromov-Wasserstein objective to align motion. It alternates between metric-alignment and patch blending to produce high-fidelity results.

Fig 2: Intuitive Concept

Fig.2: Intuitive concept of MAMM

Fig 3: FSUGW Optimization

Fig.3: FSUGW objective and alternating optimization

Waveform-to-Motion

Control periodicity and style via 1D waveforms (sine, square, center-alternating) to generate diverse motion.

Sketch-to-Motion

Draw 2D curves to specify motion trajectories. Soft and hard keyframes allow fine-grained control.

Motion-by-Numbers

Align motions using segmentation labels as one-hot vectors—no precomputed annotations required.

Motion-to-Motion

Cross-skeletal alignment enables direct mapping between different motion styles and skeletons.

Audio-to-Motion

Use MFCC features from music and speech to drive dance and gesture synthesis without training.

Music-driven Dance

Speech-driven Gestures