Skip to main content

Environment Setup

1. Quick Download​

# Clone the repository
git clone git@github.com:UBTECH-Robot/GlobalHumanoidRobotChallenge_2026_Baseline.git

# Switch to the working directory (replace workspace with your own workspace path)
cd workspace/GlobalHumanoidRobotChallenge_2026_Baseline

# Install huggingface-cli
pip install huggingface-hub

# Get simulation assets (recommended: use Git submodules)
git submodule update --init --recursive

# Or manually download assets to ./assets
# hf download UBTECH-Robotics/challenge2026_assets --local-dir ./assets --repo-type model

# Download the training dataset
hf download UBTECH-Robotics/challenge2026_dataset --local-dir ./datasets --repo-type dataset

# Download pretrained weights
hf download UBTECH-Robotics/challenge2026_baseline --local-dir ./challenge2026_baseline --repo-type model

⚠️ If you encounter network issues during download, please download the simulation assets and training dataset from the official website.

2. Build the Environment​

This project uses the Dockerfile in the repository root to build the runtime environment. The base image is nvcr.io/nvidia/isaac-sim:5.1.0. The image installs common dependencies required by Isaac Sim and copies the project into /workspace/GlobalHumanoidRobotChallenge_2026_Baseline inside the container.

2.1 Prerequisites​

  • Docker installed
  • NVIDIA Container Toolkit installed
  • NVIDIA GPU is available on the host
  • Repository code, assets/ submodule, datasets/, and challenge2026_baseline/ resources are ready

Run the following in the project root:

# Use Tsinghua mirror (recommended)
docker build --build-arg PIP_INDEX_URL=https://pypi.tuna.tsinghua.edu.cn/simple/ -t ghrc_2026:v0 -f Dockerfile . 

# Use Aliyun mirror
docker build --build-arg PIP_INDEX_URL=https://mirrors.aliyun.com/pypi/simple/ -t ghrc_2026:v0 -f Dockerfile .

⚠️ We recommend using the Tsinghua mirror command first to build the Docker image.

If you want to use a different image name, you can override it with build parameters:

# Build a Docker image
# -t image_name:tag
# .  indicates using the Dockerfile in the current directory
docker build \
    -t your-images-name:latest \
    .

2.2 Find the Keyboard evdev Path​

Before using keyboard teleoperation, you need to identify the evdev device path (/dev/input/eventX) for your keyboard. By default, the system will traverse all /dev/input/event* devices automatically.

Method 1: Find by device ID (recommended)

ls -la /dev/input/by-id/
# Find the symlink containing "keyboard", for example:
# usb-046D_C328-if01-event-kbd -> /dev/input/event2

Method 2: List all input devices

cat /proc/bus/input/devices | grep -A 3 -i keyboard
# Look for eventX in the "handlers" line

Method 3: Use evtest (interactive testing)

apt install evtest
evtest
# Select the keyboard's event device and press keys to verify

⚠️ When running inside a Docker container, make sure the input devices are properly mounted into the container.