SELF-PACED WORKSHOP

Prepare your environment

Keep the course, lab code, and fixed models in one workspace, then check the runtime.

Hands-on lab · Ubuntu 24.04 / DGX Spark

Runtime requirements

The reference environment is Ubuntu 24.04 on DGX Spark (aarch64), Python 3.12, Isaac Sim 6, and pinned Isaac Lab / Newton revisions. Simulation uses Newton’s native actuator path. Other GPUs or versions may run, but the course results do not automatically apply to them.

You need an NVIDIA driver, a GPU-enabled graphical desktop, Git, Docker, and network access to NVIDIA package sources. None of these are required for reading the website. Review the Isaac Sim license before installing the simulator.

Download the course and models

Open a terminal in the directory where you want to keep the course, then run the commands below to clone and enter the repository.

mkdir -p "$HOME/microduck"
cd "$HOME/microduck"
git clone https://github.com/NV-DLI/Microduck-Workshop.git
cd Microduck-Workshop
python3 scripts/workshop.py verify-models

The models are included in the repository, so no Hugging Face login is required. The verification command checks every model’s SHA-256. Continue only when all files match. Course weights and student outputs are stored separately.

Microduck-Workshop/
├── web/          # HTML course
├── lab/          # simulation and learning code
├── models/       # fixed course models
├── scripts/      # lab entry points
└── outputs/      # your runs (created when needed)

Prepare the simulator

First prepare system tools. The commands below install Git, Python virtual-environment support, and video encoding. Set up Docker using its Ubuntu instructions; docker info should show server information without a permission error.

sudo apt-get update
sudo apt-get install -y git python3-venv ffmpeg
docker info
export UV_HTTP_TIMEOUT=600

python3 scripts/workshop.py setup
python3 scripts/workshop.py doctor

setup prepares the pinned runtime. If you already have a matching Isaac Lab checkout, bind it using --isaaclab /path/to/IsaacLab. doctor checks Python, models, GPU, the desktop display setting, and navigation dependencies. Resolve failed checks before continuing.

Before you continue

Run the command below in a terminal at the repository root. If it specifies an output directory, use a new one to retain earlier results.

python3 scripts/workshop.py inspect-model --model models/walk.onnx

Expected result: input and output names, tensor shapes, and model identity. If the GPU is unavailable, use the troubleshooting page before proceeding. Keep the pinned physics backend and model combination.