Robot
AI Worker
GR00T N1.5Real-robot experiment
Coffee sorting demonstration.
Inference on an external RTX 5090 workstation.
NVIDIA cross-embodiment foundation model for humanoid manipulation. It takes images, language and robot state to generate continuous action sequences, using a frozen Eagle vision-language model, flow-matching action module and FLARE training objective. Real or synthetic data can be used for task-specific post-training.
Research, evaluation and robot-specific training of GR00T N1.5 robot foundation policies.
A newer version is available: GR00T N1.7 ↗
Access and availability
- Code
- Public access
- Model files
- Public access
- Training data
- Partly available
Code and training / fine-tuning
- Access
- Public access
Verified on · 2026-10-27
- Code license
- Apache-2.0
Training data
Some training datasets are public; this does not establish access to the complete training mixture.
Model files
| File | Access | Weights license |
|---|
Getting started
- Runtime and frameworks
- NVIDIA Isaac GR00TNVIDIA GPU · bfloat16
At a glance
- Model type
- Robot foundation model / Robot policy / Vision-language-action
- Backbone
- NVIDIA Eagle VLM (frozen) + DiT flow matching + FLARE objective
- Weight parameters
- 3 B
- Input
- Language instruction / RGB image frames / Robot proprioceptive state / Robot proprioception
- Output
- Continuous action chunks
- Released
- June 11, 2025
Related robots
Technical details
Flow-matching transformer predicts continuous action chunks.
Functional purpose
Action & control
Training data
CC BY 4.0
Model card
Pretraining data: internal GR-1 data, OpenXE, simulated GR-1 (DexMG), DreamGen neural trajectories, and AgiBot-Beta. https://research.nvidia.com/labs/gear/gr00t-n1_5/ ## Technical blog [GR00T N1.5 — NVIDIA](https://research.nvidia.com/labs/gear/gr00t-n1_5/) No separate official paper verified for this version. [GR00T N1](https://arxiv.org/abs/2503.14734) · 2025-03-18.
Before you use this version: Requires task- and embodiment-specific data, configuration and validation. Published examples and simulated results do not establish safe operation on a physical robot. Available weights or code do not establish plug-and-play robot compatibility. Check the required configuration, data and interfaces in the source documentation.
Submit a source or correction
