NVIDIA · robot AI model
DreamDojo
United States · 2026
Sources retrieved: 9 November 2026 (A, B, C, D, E)
- Model type
- World modelSource and note: Model typeD huggingface.co/nvidia/DreamDojo · Retrieved 9 November 2026Note. Hugging Face model card: “DreamDojo is a generalist robot world model pretrained on 44k hours of human egocentric data, showing unprecedented generalization to diverse objects and environments.”
- Inputs
- CameraSource and note: InputsA arxiv.org/html/2602.06949 · Retrieved 9 November 2026Note. The base model is “a latent video diffusion model that predicts future frames with text and conditional frame inputs”. DreamDojo adds actions: “During pretraining, we condition each latent frame on the chunked latent actions.”
- Access
- Open weights (download)Source and note: AccessB github.com/NVIDIA/DreamDojo · Retrieved 9 November 2026Note. GitHub README news: “We released all pretrained and post-trained checkpoints (2B and 14B).” The Hugging Face repository is listed as not gated.
- Base model
- Cosmos-Predict2.5Source and note: Base modelA arxiv.org/html/2602.06949 · Retrieved 9 November 2026Note. Paper: “We establish our world model on the pretrained Cosmos-Predict2.5 model”.
- Licence type
- Free, commercial useSource and note: Licence typeE nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/ · Retrieved 9 November 2026Note. The model card names the NVIDIA Open Model License for the weights (“license_name: nvidia-open-model-license”). The licence page says “Models are commercially usable.”
Specifications
8 of 14 stated 6 not stated
2 fields hidden 4 not-stated fields with a note stay ·
How to read the figuresShow guideClose
- not statedNo source
- yesStated by a source
- 0Reported as zero
- ASource
Select a source marker for source details and notes.
Release1/1 disclosed
Released6 February 2026
Source and note: ReleasedC arxiv.org/abs/2602.06949 · Retrieved 9 November 2026Note. [A] “Submitted on 6 Feb 2026”
Input and output2/4 disclosed
Model typeWorld model
Source and note: Model typeD huggingface.co/nvidia/DreamDojo · Retrieved 9 November 2026Note. Hugging Face model card: “DreamDojo is a generalist robot world model pretrained on 44k hours of human egocentric data, showing unprecedented generalization to diverse objects and environments.”
InputsCamera
Source and note: InputsA arxiv.org/html/2602.06949 · Retrieved 9 November 2026Note. The base model is “a latent video diffusion model that predicts future frames with text and conditional frame inputs”. DreamDojo adds actions: “During pretraining, we condition each latent frame on the chunked latent actions.”
What it controlsNot stated
Source and note: What it controlsB github.com/NVIDIA/DreamDojo · Retrieved 9 November 2026Note. The model generates video and does not itself command a robot: “long-horizon autoregressive generation”.
Degrees of freedom controlledNot stated
Size and speed0/2 disclosed
ParametersNot stated
Source and note: ParametersA arxiv.org/html/2602.06949 · Retrieved 9 November 2026Note. The paper gives one size per variant and no total: “a 2B model and a 14B model”.
Control rateNot stated
Source and note: Control rateC arxiv.org/abs/2602.06949 · Retrieved 9 November 2026Note. The abstract gives the speed of the generated video, not a control rate: “accelerates DreamDojo to a real-time speed of 10.81 FPS”.
Where it runs and what it is built on1/2 disclosed
On robot or cloudNot stated
Base modelCosmos-Predict2.5
Source and note: Base modelA arxiv.org/html/2602.06949 · Retrieved 9 November 2026Note. Paper: “We establish our world model on the pretrained Cosmos-Predict2.5 model”.
Openness3/3 disclosed
AccessOpen weights (download)
Source and note: AccessB github.com/NVIDIA/DreamDojo · Retrieved 9 November 2026Note. GitHub README news: “We released all pretrained and post-trained checkpoints (2B and 14B).” The Hugging Face repository is listed as not gated.
Licence documentNVIDIA Open Model License
Source and note: Licence documentD huggingface.co/nvidia/DreamDojo · Retrieved 9 November 2026Note. Weights: the Hugging Face model card has “license_name: nvidia-open-model-license”. Code: the GitHub README says the source code is released under the “Apache-2.0 license”.
Licence typeFree, commercial use
Source and note: Licence typeE nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/ · Retrieved 9 November 2026Note. The model card names the NVIDIA Open Model License for the weights (“license_name: nvidia-open-model-license”). The licence page says “Models are commercially usable.”
The developer’s own claims1/2 disclosed
Learns new tasksNot stated
Source and note: Learns new tasksB github.com/NVIDIA/DreamDojo · Retrieved 9 November 2026Note. NVIDIA describes generalization to new objects and environments, not to new tasks: the model “demonstrates strong generalization to diverse objects and environments after post-training.”
Works across robotsYes
Source and note: Works across robotsA arxiv.org/html/2602.06949 · Retrieved 9 November 2026Note. Paper: “To adapt DreamDojo to different embodiments, we reset the action conditioning layer and learn the new action space through finetuning.” The Hugging Face repository holds post-trained checkpoints for several robot setups (“14B_AgiBot_post-train”, “14B_G1_post-train”, “14B_GR1_post-train”, “14B_YAM_post-train”).Yes only if the developer itself says the model runs on different robot models, for example robots from two manufacturers or two of one manufacturer’s models. Several units of the same robot do not count.
Notes
NVIDIA describes DreamDojo as an interactive world model that learns from large-scale human videos (GitHub README).
The pretraining data is 44k hours of egocentric human video.
NVIDIA released pretrained checkpoints and post-trained checkpoints for the G1, GR-1, AgiBot and YAM robot setups.
Sources
One line per source, not per figure: a source has one retrieval date.
- Aarxiv.org/html/2602.06949Retrieved 9 November 2026
- Bgithub.com/NVIDIA/DreamDojoRetrieved 9 November 2026
- Carxiv.org/abs/2602.06949Retrieved 9 November 2026
- Dhuggingface.co/nvidia/DreamDojoRetrieved 9 November 2026
- Envidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/Main source. Retrieved 9 November 2026
More models
Same developer
Cosmos 3NVIDIA
Cosmos PolicyNVIDIA
Cosmos Reason 2NVIDIA
DreamZeroNVIDIA
- No photograph
GR00T 2NVIDIA
- No photograph
Isaac GR00T N1.7NVIDIA
