HDMI: Human-to-Motion Intelligence for the Unitree G1
Learn how to teach the Unitree G1 humanoid robot to move by learning directly from human videos. This course covers the complete Human-to-Motion Intelligence (HDMI) pipeline — converting video into motion datasets, visualizing and validating humanoid motion, training learned control policies, and deploying those behaviors onto a real Unitree G1 humanoid robot.
Start LearningThis course includes
- Human motion extraction from video
- G1 motion dataset creation & visualization
- Training datasets into motion policies
- Deploying learned behaviors to a real G1
About this course
This course teaches you how to create humanoid robot behaviors for the Unitree G1
by learning directly from human motion captured in video. Instead of hand-designing
controllers or tuning rewards from scratch, you will build a Human-to-Motion
Intelligence (HDMI) pipeline that converts video into structured motion datasets,
visualizes and validates those motions in simulation, trains learned control
policies, and deploys them onto a real Unitree G1 humanoid.
The course emphasizes end-to-end embodied AI workflows used in modern humanoid
research, bridging computer vision, motion representation, simulation, and
real-world robot deployment.
Skills you'll gain
- Understanding the Human-to-Motion Intelligence (HDMI) pipeline
- Extracting human motion from monocular video
- Mapping human kinematics to the Unitree G1 joint structure
- Creating robot-ready motion datasets from video
- Cleaning, aligning, and validating motion data
- Visualizing humanoid motion in simulation environments
- Training motion datasets into learned control policies
- Evaluating motion stability and physical feasibility
- Exporting trained policies for deployment
- Deploying learned behaviors to a real Unitree G1 humanoid
- Iterative refinement and sim-to-real motion tuning
