1 of 30

Interactive Character Control with

Auto-Regressive Motion Diffusion Models

Yi Shi 1, 2 Jingbo Wang 2 Xuekun Jiang 2 Bingkun Lin 3

Bo Dai 2 Xue Bin Peng 1,4

Vi

1

2

4

3

© 2024 SIGGRAPH. ALL RIGHTS RESERVED.

2 of 30

Related Work

Human Motion Diffusion Model (MDM)

[Tevet et al. 2022]

Space-Time Models

Space-Time

Model

Noise

3 of 30

Related Work

Auto-Regressive Models

MotionVAE

[Hung et al. 2021]

Auto-Regressive Model

4 of 30

Auto-Regressive Motion Diffusion Model

Random Sampling

A-MDM:

Hierarchical Reinforcement Learning

Conditional Inpainting

5 of 30

Overview

Stage I : Base A-MDM Training

Stage II : Task-based Control

6 of 30

AMDM Base Model

A-MDM

7 of 30

AMDM Base Model: Training

Denoiser

Loss

Step

8 of 30

AMDM Base Model: Runtime

Denoiser

Step

T

x

9 of 30

A-MDM Denoising

10 of 30

A-MDM Denoising

11 of 30

MLP based Network Architecture

A-MDM

MLP

12 of 30

Denoising Steps

13 of 30

Denoising Steps

14 of 30

Random Synthesis - LAFAN1

15 of 30

Random Synthesis - LAFAN1

16 of 30

Random Synthesis - HumanML3D

17 of 30

Random Synthesis - HumanML3D

18 of 30

Overview

Stage I : Base A-MDM Training

Stage II : Task-based Control

Conditional Inpainting

19 of 30

Spatial Inpainting

A-MDM

Denoiser

Denoiser

Denoiser

User Command

20 of 30

Spatial Inpainting

21 of 30

Keyframe Motion In-betweening

Source Motion

Target Motion

22 of 30

Keyframe Motion In-betweening

Time

(Frames)

Denoising Step

Source Motion

Target Motion

Source Motion

23 of 30

Keyframe Motion In-betweening

24 of 30

Overview

Stage I : Base A-MDM Training

Hierarchical RL

Stage II : Task-base Control

25 of 30

Hierarchical Reinforcement Learning

Action

A-MDM

Controller Policy

26 of 30

Hierarchical Reinforcement Learning

A-MDM

Controller Policy

Action

Denoiser

Denoiser

Denoiser

27 of 30

Hierarchical Reinforcement Learning

28 of 30

29 of 30

Hierarchical Reinforcement Learning

30 of 30

Hierarchical Reinforcement Learning