Installation

git clone https://github.com/michelebri/human_to_robot --recurse-submodules
cd human_to_robot

conda create -n <name_env> python=3.10
conda activate <name_env>

conda install -c conda-forge chumpy
pip install -r requirements.txt

pip install -e GVHMR
pip install -e GMR

Configuration

config.yaml must contain at least:

output_path: /abs/path/to/output_root
robot: booster_t1

Notes:

• video_path is no longer read from YAML.
• Input source is required from CLI.
• Output path is normalized to include robot name if missing. Example: /abs/path/to/output_root/booster_t1.

Run

Use video_to_robot.py for both single and batch use.

Single video

python video_to_robot.py input_video/video.mp4 --config config.yaml

Folder (batch)

python video_to_robot.py input_video --config config.yaml

Behavior:

• Supported extensions: .mp4, .avi, .mov, .mkv, .webm.
• Skips a video only when both final files exist and non-verbose generated artifacts are complete:
  • <video_name>.pkl
  • <video_name>_joint_poses.npy
• Without --rewrite, videos with missing non-verbose artifacts are fully regenerated; complete videos are skipped.
• Add --rewrite to force full regeneration for all videos.

Output Files

For each processed video <video_name>, outputs are under:

• <output_root>/<robot>/<video_name>/<video_name>.pkl
• <output_root>/<robot>/<video_name>/<video_name>_joint_poses.npy

Depending on settings, you can also get:

• <output_root>/<robot>/<video_name>/<video_name>_robot_motion.mp4
• <output_root>/mujoco_csv/<robot>/<video_name>_*.csv (if converter exists for the selected robot)

The mujoco_csv/<robot> directory is auto-created if missing.