Course navigation
Week 5: VLA ArchitecturesDay 31
GR00T N1.5 / N1.6 architecture
This is a valid v1.0 placeholder page for the later curriculum arc. Full interactive lab treatment ships after Week 1 dogfooding.
LECTURE & READING
Glossary primer (10 min)
- GR00T — NVIDIA's humanoid Modern Robot LearningFoundation modelA large pretrained model that can be adapted to many tasks. project, launched at GTC 2024.
- GR00T N1 — First public release (Mar 2025). 2B params. Diffusion Core ConceptsActionA command the robot sends to its motors, controller, or low-level system. head.
- GR00T N1.5 — Updated 2025. Modern Robot LearningSkillA reusable behavior like grasp, push, place, or open drawer. tokens (discrete high-level intents).
- GR00T N1.6 — Public release at GTC 2026 (~Apr 2026). DINOv3 vision; world-model-aware loss; 3B params.
- GR00T N1.7 EA — Early-access only as of GTC 2026. Internal docs reference but not public.
- Eagle-2 — NVIDIA's Modern Robot LearningVision-Language Model (VLM)A model that understands both images and text. backbone. 1.4B params.
- Modern Robot LearningSkillA reusable behavior like grasp, push, place, or open drawer. token — Discrete embedding for an abstract sub-task ("grasp", "rotate", "transfer"). Allows hierarchical Imitation & Reinforcement LearningReinforcement Learning (RL)Teaching a robot through trial and error using rewards..
- Humanoid embodiments — Unitree G1, Boston Movement, Mechanics & Robot BodyDynamicsThe study of motion including forces, torques, mass, and inertia. Atlas (limited), Unitree H1, 1X Neo, Apptronik Apollo.
Real-world analogy
GR00T is the "industrial fleet operating system": built specifically for humanoids, supports multiple makes (G1, H1, Atlas, Neo), comes with NVIDIA's full tooling stack (Isaac Sim, Cosmos Modern Robot LearningWorld modelA model that predicts how the world will change after actions., etc.).
Hour 1 — Reading
- GR00T N1.6 release blog (~20 min): https://www.nvidia.com/en-us/research/groot/
- GR00T-Mimic paper (related; Imitation & Reinforcement LearningTeleoperation (teleop)A human remotely controlling the robot, often to collect demonstrations. data scaling) — abstract + Sec 3 (~20 min): https://arxiv.org/abs/2410.24221
Hour 2 — Read GR00T's open-source code
git clone https://github.com/NVIDIA/Isaac-GR00T
cd Isaac-GR00T
# Read in this order (~30 min):
# - gr00t/models/gr00t_n1.py (architecture)
# - gr00t/data/dataset.py (data format)
# - examples/ (inference and fine-tune scripts)LAB
Hour 3 — Lab: GR00T N1 inference + LoRA on humanoid sim data (75 min)
What you're building. Run GR00T N1 (open-weights) Robot LearningInferenceUsing a trained model to make predictions or choose actions. on a humanoid Core ConceptsTaskThe job the robot is supposed to complete, such as pick-and-place, navigation, or drawer opening., then LoRA-fine-tune on 50 episodes of simulated humanoid Manipulation & TasksManipulationUsing a robot arm or hand to move or interact with objects..
Step 1 — Install (15 min)
cd ~/robo47-il/Isaac-GR00T
uv pip install -e .
huggingface-cli download nvidia/GR00T-N1-2B --local-dir checkpoints/gr00t_n1Step 2 — Zero-shot eval on humanoid task (30 min)
Full source continues in the committed curriculum files. The v1.0 page exposes the day flow and lab surface without inventing content.