Installation

Check the host

The supported baseline is Linux x86_64, Python 3.11, GLIBC 2.35 or newer, Isaac Sim 5.1.0, and the Isaac Lab revision pinned by setup_env.sh. Use an NVIDIA RTX-capable GPU with enough VRAM for Isaac Sim, at least 32 GB host RAM, and at least 50 GB free SSD space for the simulator.

For desktop, streaming, or camera rendering, use NVIDIA’s validated Linux driver 580.65.06 (R580) for Isaac Sim 5.1.0. Check the installed version:

nvidia-smi --query-gpu=name,driver_version,memory.total --format=csv

See NVIDIA’s version-specific requirements before configuring the host. A newer driver is not automatically compatible with this release. In particular, NVIDIA documents an RTX SceneDB startup crash with newer driver branches; troubleshooting links the upstream diagnosis. A GPU visible to nvidia-smi alone does not verify camera or viewport rendering.

Install Git, Git LFS, GCC, G++, Make, and uv. On Ubuntu, build-essential provides GCC, G++, and Make. Install ffmpeg if you intend to record videos.

Install the simulation environment

Run from the repository root:

git clone https://github.com/analogdevicesinc/dex-01-sim-assets.git
cd dex-01-sim-assets
git lfs install
git lfs pull
./setup_env.sh
source ../env_isaaclab/bin/activate

The setup script installs Python 3.11, Isaac Sim 5.1.0, PyTorch, Isaac Lab’s extensions, and this package. It installs CMake 3.x for native dependencies such as egl-probe. uv fetches Python, so your host Python need not be 3.11.

A fresh Isaac Lab checkout is pinned to b0542fe2d45bf91c4e1d9ef6952b9c709c80b4e8. Existing checkouts and environments are reused. Read the reported versions before reusing them; the script does not replace a working installed simulator just because its version differs. The first simulator launch can request NVIDIA’s EULA, download extensions, and compile shaders.

Choose installation locations

Use absolute paths for custom locations:

./setup_env.sh --venv /path/to/dex01-env --isaaclab /path/to/IsaacLab
source /path/to/dex01-env/bin/activate

With an existing compatible Isaac Lab installation, fetch the LFS assets and install this package into its Python environment:

/path/to/IsaacLab/isaaclab.sh -p -m pip install -e source/dex01_sim_asset

All example commands run from the repository root in that environment. Simulator modules must be imported after AppLauncher starts Isaac Sim.

Confirm the installation

python - <<'PY'
from importlib.metadata import version
print("Isaac Sim:", version("isaacsim"))
print("PyTorch:", version("torch"))
PY
git -C ../IsaacLab rev-parse HEAD
MPLBACKEND=Agg python scripts/dex01_on_tesollo_dg5f.py \
  --headless \
  --flat-indenter \
  --max-steps 240

For a custom checkout, replace ../IsaacLab with the path passed to setup. The headless smoke check prints per-finger forces, runs 240 steps, and exits. Finger 2 should settle near 0.981 N while the other fingers remain unloaded. This local cube fixture does not download the gear asset. Continue with your first visual simulation and then complete physical-force validation.

View a remote simulation

After installing and activating the environment on the simulation host, run from the repository root:

MPLBACKEND=Agg python scripts/dex01_on_tesollo_dg5f.py --livestream=2

Use the RTX-capable GPU and validated R580 driver described above. On your viewing computer, install the Isaac Sim WebRTC Streaming Client. Wait for Setup complete..., enter the simulation host’s reachable IP address, and connect. TCP 49100 and UDP 47998 must be reachable from that computer. Run one streaming simulator at a time on these ports. --livestream=2 is for a private network; NVIDIA’s guide covers public endpoints.

The streamed viewport includes the hand, tactile frames, and normal loads in newtons. Only environment 0 is displayed in multi-environment runs. The same overlay is available in simple_example.py and verify_dex01_sim.py. Use --no-tactile-ui to hide it or --debug-vis for 3D sampling markers. Stop with Ctrl+C. Pure headless runs without streaming skip the viewport. Continue with your first simulation to interpret the result.