Robot foundation models, zero to hero · Pilot
NVIDIA GR00T Practical Course
A hands-on course that starts at zero and ends with you teaching NVIDIA's open GR00T model a new robot task, testing it honestly, and planning how to run it on a real robot. Every lesson has work to do on a cloud GPU lab.
- Modules
- 13
- Level
- Beginner to master
- Model
- NVIDIA GR00T N1.7
- Labs
- Cloud GPU machines
What you will learn
-
1
Start Here
What a robot foundation model is, how the course works, and your first session on a GPU lab.
-
2
The Terminal and the Lab GPU
Files, commands, SSH, tmux and git: the everyday tools of anyone who trains robot models.
-
3
Python for Robot Data
Python from the ground up, then NumPy, tables and plots, all on real robot recordings.
-
4
How Robots Move and Sense
Joints, states, actions, cameras and the control loop, the words every robot model speaks.
-
5
How Machines Learn
Data, loss and training, honest testing, GPUs and PyTorch, and the transformers inside GR00T.
-
6
Learning from Demonstrations
Imitation learning, why small errors grow, action chunks, diffusion and vision-language-action models.
-
7
Robot Datasets
The LeRobot format, recording good demonstrations, checking a dataset and writing modality.json.
-
8
Running GR00T
Install GR00T, call it from Python, measure it on recorded data, and serve it to a robot over the network.
-
9
Fine-Tuning GR00T
Teach GR00T a new robot: the modality config, the training run, checkpoints and honest results.
-
10
Simulation and Closed-Loop Testing
Test a policy that really acts, in Isaac Lab and standard benchmarks, and report a success rate you can trust.
-
11
More Data: Synthetic and Generated
Grow a few human demonstrations into many with Isaac Lab Mimic, and meet world models that dream new data.
-
12
Deploying to Real Robots
Speed, Jetson and TensorRT, safety, and how humanoids split thinking from balance.
-
13
Capstone: From Data to a Working Robot
One project from start to finish: plan it, collect data, fine-tune, test and defend your results.
How the course works
Built for beginners
You need no robotics, AI or Linux experience. Every new word is explained the first time it appears.
A GPU lab for every lesson
Labs run on cloud GPU machines that are ready to use. You do not need a powerful computer.
Practice on every screen
Each lesson is a few short screens, each with something to do, then a lab, a quiz and a skills check.
Honest results
From the first model you train, you learn to test on unseen data and to report how sure you are.