Free hands-on learning for developers

Learn NVIDIA
Physical AI by doing.

Explore robotics, physics engines, world models, and autonomous driving through NVIDIA tools. Start with a recorded experiment or browser lab, then use the code and guides to run your own.

GR00T puts the cube in the bowlIsaac Sim GUI · RTX 5090

Four ways into Physical AI

See the result, understand the inputs and outputs, then choose a lab. Each field shows what you can try now and what needs a GPU, model, or dataset.

Watch four fields in action

Robotics

Judge a GR00T pick-and-place trial, explore obstacle avoidance, and follow a Microduck walking policy.

  • Isaac Lab
  • GR00T
  • cuRobo

Browser lab · recorded runs and code

Explore this field

Physics engines

Compare collision shapes, friction, and bouncing in Isaac Sim; inspect Newton and Warp simulation results.

  • Isaac Sim
  • PhysX
  • Newton

Recorded runs · scripts and USD scenes

Explore this field

3D worlds & world models

Generate a 3D world with WorldGen, explore fVDB reconstruction and procedural houses, then compare Marble, Lyra and video models.

  • WorldGen
  • Lyra
  • fVDB

Interactive 3DGS · Recorded experiments · generation and reconstruction separated

Explore this field

Autonomous driving

See city, rain, sunset and fog from real CARLA ego cameras. Follow with Alpamayo inference and AlpaSim policy evaluation.

  • CARLA
  • AlpaSim
  • NuRec

CARLA recordings · source, sensor poses and checks

Explore this field
Also explore CAD, 3D assets, and synthetic data

Choose what you want to build

Each path takes you from the first step to a result you can check.

How the tools work
CAD workbench open in the Isaac Sim viewport

Browser → Python → RTX

From CAD to camera data

Build a workbench scene and generate aligned RGB, depth, and segmentation data.

What you’ll produce
A USD scene, a reconstructed point cloud, and verified camera data.
View path
Robot motion

Browser → RTX

Robot policies and motion

Score a GR00T run, check an obstacle-aware path, and train a walking policy.

What you’ll produce
Episode outcomes, planned paths, and a training checkpoint.
View path
Four environments from the driver’s seat

Recorded cameras → inference → evaluation

Ego cameras to driving evaluation

Start with real CARLA camera recordings, inspect Alpamayo outputs, then follow AlpaSim policy evaluation.

What you’ll produce
Camera frames and trajectories first; policy rollout metrics in the AlpaSim extension.
View path

Find an example in your field

Compare results once, then pick a scene or tool to try on your own.

Run the examples yourself

Use the same scripts, scenes, and saved outputs shown in the labs. If you’re new to NVIDIA tools, start with the setup guides.

Find the right tools for your own project

When you have a scene, driving log, or robot demonstration, use the map to find what you need for the next step.

Ecosystem map

Start with the input you have. The map traces how 3D assets, driving logs, and robot demonstrations each lead to a result you can evaluate.

Open full-size mapThree workflows. Build an OpenUSD scene from CAD and generate labeled data in Isaac Sim. Reconstruct driving logs with NuRec and evaluate a policy in AlpaSim. Turn demonstrations into training data for Isaac Lab or GR00T, then evaluate and deploy the policy.

See how the tools fit together

Choose a goal and trace it through the datasets, catalog, guides, and result checks.

Build a scene

  1. Data and assets

    Workbench CAD

    Start with a STEP file and the real tabletop dimensions.

    Browse 3D assets
  2. Catalog

    AIF Pipeline Samples

    Convert the STEP file to an OpenUSD asset.

    Find AIF in the catalog
  3. Hands-on guide

    OpenUSD → Isaac Sim

    Bring in the workbench and add a collider to the tabletop.

    Follow the CAD guide
  4. Result check

    A cube rests on the table

    Check the tabletop dimensions and watch the cube settle on it.

    Run the collision check

Evaluate driving

  1. Data and assets

    Driving data

    Start with a sample scene from the NuRec driving dataset.

    Find the driving dataset
  2. Catalog

    NuRec

    Reconstruct the driving scene so it can be rendered from new viewpoints.

    Find NuRec in the catalog
  3. Hands-on guide

    AlpaSim

    Run a baseline driving policy in the reconstructed scene.

    Follow the AlpaSim guide
  4. Result check

    Collision and off-road metrics

    Review the driving video and metrics before comparing policies.

    Find the recorded metrics

Test a robot policy

  1. Data and assets

    Robot demonstrations

    Check what the recorded cameras, observations, and actions contain.

    Browse demonstration datasets
  2. Catalog

    GR00T checkpoint

    Check whether the checkpoint supports your robot or benchmark.

    Compare GR00T checkpoints
  3. Hands-on guide

    Robot model guide

    Use the official example to check input and action formats before running the policy.

    Follow the model guide
  4. Result check

    A baseline run

    Record the success criteria and initial conditions to compare later runs.

    Open the evaluation checklist