Installation
Route 1: Native Execution
- Install ROS 2 natively on your system.
- Set up the ROS 2 context and add a ROS 2 subscriber block.
- Ensure Fast DDS is properly configured for ROS 2 communication.
- Clone and integrate the following workspace repository: isaac-sim/IsaacSim-ros_workspaces.
Additional Notes:
- For headless installation, refer to the Isaac Sim headless container guide.
- For development and deployment, explore Isaac Sim’s start-dev and deployment workflows. A helpful walkthrough is available here.
- See the official ROS bridge documentation for integration details.
- verify GPU access inside Docker:
1
2
docker run --rm --runtime=nvidia --gpus all \
nvcr.io/nvidia/cuda:12.8.0-base-ubuntu24.04 nvidia-smi
2. Pull Isaac Sim container
1
docker pull nvcr.io/nvidia/isaac-sim:6.0.0
NVIDIA’s current container docs use nvcr.io/nvidia/isaac-sim:6.0.0. (Isaac Sim Documentation)
3. Create persistent cache folders
Do this once:
1
2
3
4
5
6
7
8
9
10
11
12
13
mkdir -p ~/docker/isaac-sim/cache/main
mkdir -p ~/docker/isaac-sim/cache/computecache
mkdir -p ~/docker/isaac-sim/config
mkdir -p ~/docker/isaac-sim/data
mkdir -p ~/docker/isaac-sim/logs
mkdir -p ~/docker/isaac-sim/pkg
mkdir -p ~/.cache/ov/hub
mkdir -p ~/isaac_projects/caliper_sdg
mkdir -p ~/isaac_projects/assets
mkdir -p ~/isaac_projects/output
sudo chown -R 1234:1234 ~/docker/isaac-sim ~/.cache/ov/hub ~/isaac_projects
The cache mounts matter because first launch compiles/warmups lots of shaders; keeping the cache makes later runs much faster. NVIDIA’s docs also warn that stale cache mounts can cause config or launch problems, in which case removing/recreating ~/docker/isaac-sim can help. (Isaac Sim Documentation)
5. Run compatibility checker inside container
1
2
3
4
5
6
7
8
9
10
11
12
13
docker run --entrypoint bash -it --gpus all --rm --network=host \
-e "ACCEPT_EULA=Y" \
-v ~/docker/isaac-sim/cache/main:/isaac-sim/.cache:rw \
-v ~/docker/isaac-sim/cache/computecache:/isaac-sim/.nv/ComputeCache:rw \
-v ~/docker/isaac-sim/logs:/isaac-sim/.nvidia-omniverse/logs:rw \
-v ~/docker/isaac-sim/config:/isaac-sim/.nvidia-omniverse/config:rw \
-v ~/docker/isaac-sim/data:/isaac-sim/.local/share/ov/data:rw \
-v ~/docker/isaac-sim/pkg:/isaac-sim/.local/share/ov/pkg:rw \
-v ~/.cache/ov/hub:/var/cache/hub:rw \
-v ~/isaac_projects:/workspace:rw \
-u 1234:1234 \
nvcr.io/nvidia/isaac-sim:6.0.0 \
./isaac-sim.compatibility_check.sh --/app/quitAfter=10 --no-window
You want to see something like:
1
System checking result: PASSED
NVIDIA’s docs explicitly mention this expected pass message. (Isaac Sim Documentation)
Note: ACCEPT_EULA=Y is required to accept the container license. PRIVACY_CONSENT=Y is optional telemetry opt-in; I would omit it unless you intentionally want to opt in. NVIDIA documents both environment variables. (Isaac Sim Documentation)
Here in my setup I encountered:
1
2
3
GPU 0: VRAM [not enough]
total: 8.59 GB
minimum: 10 GB
But I expected potential crashes and tried this anyways:
1
2
3
4
5
6
7
8
9
10
11
12
docker run --name isaac-sim --entrypoint bash -it --gpus all --rm --network=host \
-e "ACCEPT_EULA=Y" \
-v ~/docker/isaac-sim/cache/main:/isaac-sim/.cache:rw \
-v ~/docker/isaac-sim/cache/computecache:/isaac-sim/.nv/ComputeCache:rw \
-v ~/docker/isaac-sim/logs:/isaac-sim/.nvidia-omniverse/logs:rw \
-v ~/docker/isaac-sim/config:/isaac-sim/.nvidia-omniverse/config:rw \
-v ~/docker/isaac-sim/data:/isaac-sim/.local/share/ov/data:rw \
-v ~/docker/isaac-sim/pkg:/isaac-sim/.local/share/ov/pkg:rw \
-v ~/.cache/ov/hub:/var/cache/hub:rw \
-v ~/isaac_projects:/workspace:rw \
-u 1234:1234 \
nvcr.io/nvidia/isaac-sim:6.0.0
In the container,
1
2
cd /isaac-sim
./runheadless.sh -v
Until you see Isaac Sim Full Streaming App is loaded. Then open Isaac Sim WebRTC Streaming Client on the same machine and connect to: 127.0.0.1
If connecting from another laptop over LAN/Tailscale, use your host IP instead:
Then, download isaac sim webrtc: https://docs.isaacsim.omniverse.nvidia.com/6.0.1/installation/download.html#isaac-sim-latest-release, and start a webrtc client
Concepts
USD: Universal Scene Description is the language Isaac Sim uses to describe the robot and its environment. A USD scene can contain the robot, objects, lights, cameras, materials, transforms, physics properties, sensors, and scene hierarchy. It’s basically a scene graph + asset format:
1
2
3
4
5
6
7
8
9
World
├── Robot
│ ├── base_link
│ ├── arm_joint_1
│ └── camera_sensor
├── Table
├── Object
├── Lights
└── Physics settings
A .usd file is not just a mesh like .obj or .stl. It can store or reference many types of scene information, including geometry, materialuiis, lighting, animation, and physics schemas. OpenUSD describes USD as a system for authoring, composing, and reading hierarchically organized 3D scene descriptions. In Isaac Sim, you use USD files to:
- load robots and environments
- import CAD or mesh assets
- define object transforms and hierarchy
- attach materials and textures
- set collision and rigid-body properties
- compose scenes from multiple referenced assets
- reuse the same robot or object across simulations
Gotchas
- The runscript creates several cache directories, like warp, ComputeCache, etc. but it doesn’t mount /isaac-sim/.cache/ov/texturecache.
- The container can’t create that directory since uid 1234 doesn’t own the base path inside the container
- The simulation still runs; it just won’t cache textures between sessions, meaning slightly slower texture loading on first use.
1
2
3
2026-06-24T12:41:34Z [120,073ms] [Error] [omni.rtx] ResourceManager: Failed to create the texture cache
2026-06-24T12:41:34Z [120,073ms] [Error] [omni.rtx] Failed to create texture cache in /isaac-sim/.cache/ov/texturecache
2026-06-24T12:41:34Z [120,073ms] [Error] [omni.rtx] Failed to create texture cache folder /isaac-sim/.cache/ov/texturecache
- You can run the above command
docker run --name isaac-sim --entrypoint bashif the socket is shared with the host. The container talks directly to the host’s daemon. The key line is indocker-compose.yml
1
- /var/run/docker.sock:/var/run/docker.sock
IsaacSim vs Blender
| Category | Isaac Sim | Blender |
| ————————— | ——————————————————————— | ———————————————————————————————————————– |
| Main purpose | Robotics simulation, synthetic data, robot testing | 3D modeling, animation, rendering, asset creation |
| Synthetic labels | Built-in Replicator annotators: RGB, bbox, segmentation, depth, etc. | Possible, but you usually build/export scripts yourself |
| Robotics physics | Built around robotics sim, PhysX/Newton, robot import, sensors, ROS 2 | Has physics, but not robotics-stack oriented |
| USD support | Native foundation; Isaac Sim scene language is OpenUSD | Can import/export USD, but Blender docs note some USD composition features like layers/references are not fully handled |
| Materials | USD/MDL/Omniverse materials, RTX renderer | Excellent materials in Blender/Cycles/Eevee, but translation to USD/Isaac may be imperfect |
| Rendering | RTX/Omniverse renderer, good for synthetic robotics data | Cycles is excellent for photorealistic rendering |
| Headless dataset generation | Strong | Also possible with Python CLI, but labels require more custom work |
| Learning curve | Heavy, more brittle, GPU/driver/container issues | Easier to use interactively, great community/tools |
| Best file format | USD/USDZ | .blend, but can import/export many formats including USD/OBJ/FBX/STL |
Blender can render beautiful images, but if you want object detector labels, you usually need extra scripting. Blender is better for asset preparation.
- fix mesh normals
- decimate high-poly CAD
- repair bad geometry
- assign UVs
- make textures
- make nicer materials
- create background rooms
- create clutter objects
- convert/export OBJ/FBX/STL/USD
IsaacSim is also better at randomizing things like camera pose, lighting, etc.
