mancub/NitroGen

1---2datasets:3- nvidia/NitroGen4tags:5- behavior6- cloning7- gaming8- agent9---10<img src="https://cdn-uploads.huggingface.co/production/uploads/67d8509cb6b70254852d734d/u3VY6_KoT6tEs86YPehU2.gif" width="100%" />11 12<div align="center">13  <p style="font-size: 1.2em;">14    <a href="https://nitrogen.minedojo.org/"><strong>Website</strong></a> | 15    <a href="https://huggingface.co/nvidia/NitroGen"><strong>Model</strong></a> |16    <a href="https://huggingface.co/datasets/nvidia/NitroGen"><strong>Dataset</strong></a> |17    <a href="https://nitrogen.minedojo.org/assets/documents/nitrogen.pdf"><strong>Paper</strong></a>18  </p>19</div>20 21# Model Overview22 23### Description:24 25NitroGen is a unified vision-to-action model designed to play video games directly from raw frames. It takes video game footage as input and outputs gamepad actions. Unlike models trained with rewards or task objectives, NitroGen is trained purely through large-scale imitation learning on videos of human gameplay. NitroGen works best on games designed for gamepad controls (e.g., action, platformer, and racing games) and is less effective on games that rely heavily on mouse and keyboard (e.g., RTS, MOBA).26 27The goal of the NitroGen project is to explore whether large-scale training on diverse human gameplay leads to emergent, general-purpose embodied abilities, similar to how scaling has unlocked emergent behaviors in large language models.28 29Potential applications include next-generation game AI, automated QA for video games, and advancing research in general embodied AI.30 31NitroGen 1 was developed by NVIDIA and is the first model of the series. This model is for research and development only.32 33### License/Terms of Use:34 35Governing Terms:  [NVIDIA License](https://developer.download.nvidia.com/licenses/NVIDIA-OneWay-Noncommercial-License-22Mar2022.pdf).36 37Additional Information: [Apache License](https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/apache-2.0.md) for [https://huggingface.co/google/siglip2-base-patch16-224]().38 39### Deployment Geography:40Global <br>41 42### Use Case: <br>43Researchers, engineers, open source community, companies, gamers. Potential applications include next-generation game AI, automated testing for video games, and generally advancing research in embodied AI.<br>44 45### Release Date:  <br>46GitHub 12/19/2025 via []() <br>47GitHub 12/19/2025 via [https://huggingface.co/nvidia/NitroGen](https://huggingface.co/nvidia/NitroGen) <br> 48 49## References:50[VPT](https://arxiv.org/abs/2206.11795), a Minecraft agent trained from internet videos.51[SIMA](https://arxiv.org/abs/2404.10179), a multi-game agent trained to follow text instructions.52[GR00T N1](https://arxiv.org/abs/2503.14734), an open foundation model for generalist humanoid robots.53<br> 54 55## Model Architecture:56**Architecture Type:** Vision Transformer, Diffusion Transformer <br>57 58**Network Architecture:** 59- RGB frames are processed through a pre-trained vision transformer (SigLip2).60- A diffusion matching transformer (DiT) then generates actions, conditioned on SigLip output.61<br>62 63**This model was developed based on** SigLip2 <br> 64 65**Number of model parameters:** $4.93 × 10^8$ <br>66 67## Input(s): <br>68**Input Type(s):** Image <br>69 70**Input Format(s):** Red, Green, Blue (RGB) <br>71 72**Input Parameters:** Two-Dimensional (2D) <br>73 74**Other Properties Related to Input:** 256x256 Images75 76## Output(s)77 78**Output Type(s):** Actions for gamepad/game controllers<br>79 80**Output Format(s):** Tabular <br>81 82**Output Parameters:** 2D: one action dimension and one temporal dimension <br>83 84**Other Properties Related to Output:** The output has shape 21x16, two 2D Continuous-value vectors for each joystick, 17 binary values for each button.85 86 87Our AI models are designed and/or optimized to run on NVIDIA GPU-accelerated systems. By leveraging NVIDIA’s hardware (e.g. GPU cores) and software frameworks (e.g., CUDA libraries), the model achieves faster training and inference times compared to CPU-only solutions. <br> 88 89## Software Integration:90**Runtime Engine(s):** 91No runtime engine was used.92 93 94**Supported Hardware Microarchitecture Compatibility:** <br>95* NVIDIA Blackwell <br>96* NVIDIA Hopper<br>97 98**Preferred/Supported Operating System(s):**99 100The integration of foundation and fine-tuned models into AI systems requires additional testing using use-case-specific data to ensure safe and effective deployment. Following the V-model methodology, iterative testing and validation at both unit and system levels are essential to mitigate risks, meet technical and functional requirements, and ensure compliance with safety and ethical standards before deployment. <br>101 102* Linux <br>103* Windows <br>104 105## Model Version(s):106V1  <br>107 108 109## Training, Testing, and Evaluation Datasets:110 111## Training Dataset:112 113**Data Modality**<br>114* Image <br>115* Video <br>116 117**Image Training Data Size**<br>118* More than 1 Billion Images <br>119 120 121**Video Training Data Size**<br>122* 10,000 to 1 Million Hours <br>123 124**Data Collection Method by dataset** <br>125* Automated  <br>126 127**Labeling Method by dataset** <br>128* Synthetic <br>129 130**Properties:** 40,000 publicly available videos, labeled with frame-wise actions <br>131 132### Testing Dataset:133 134**Data Collection Method by dataset** <br>135* Automated  <br>136 137**Labeling Method by dataset** <br>138* Synthetic <br>139 140**Properties:** 40,000 publicly available videos, labeled with frame-wise actions <br>141 142### Evaluation Dataset:143 144**Data Collection Method by dataset** <br>145* Automated  <br>146 147**Labeling Method by dataset** <br>148* Synthetic <br>149 150**Properties:** 40,000 publicly available videos, labeled with frame-wise actions <br>151 152# Inference:153**Acceleration Engine:** None <br>154**Test Hardware:** H100 <br>155 156## Ethical Considerations:157NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications.  When downloaded or used in accordance with our terms of service, developers should work with their internal model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse. <br> 158 159For more detailed information on ethical considerations for this model, please see the Model Card++ Bias, Explainability, Safety & Security, and Privacy Subcards.160 161Please report model quality, risk, security vulnerabilities or NVIDIA AI Concerns [here](https://www.nvidia.com/en-us/support/submit-security-vulnerability/).