ViT-H · Method Overview
Back to project ↗Follow the video-to-humanoid pipeline
Pre-train
Humanoid Motion Generation from Videos
Input
video

Video condition set
Camera
orientation
BBox &
keypoints
Image
feature



Diffusion-based
motion generator
Direct
G1 motion
prediction

Preceding
segment
segment
(AR generator
only)
only)
Post-train
Post-Training with Motion Tracking Feedback
Motion generation
Motion
generator



B
A
Candidates






Motion tracking
A
B
Tracking targets
Deployable tracker






A
B
Privileged tracker






R
local
R
succ
R
global
R
succ
Preference optimization
More
likely
A
B
Less
likely






DPO update
Deploy
Real-World Deployment
Offline generation · Video-to-Humanoid

Motion
tracker
Motion
generator

Online generation · Video teleoperation


Locomotion teleoperation
Loco-manipulation teleoperation
Motion
tracker
AR motion
generator
Motion
tracker
AR motion
generator


Preceding segment
Preceding segment