NVIDIA Releases Cosmos3-Nano Policy Model for LIBERO-10 Environment
nvidia/cosmos3-nano-libero10-sdg-export-D2
NVIDIA has released a trained policy model for the Cosmos3-Nano framework, specifically adapted for the LIBERO-10 environment. The model was trained on a combination of real and synthetic data, achieving success rates of 93.9% in standard scenes and 84.7% in perturbed scenes. The model is part of a broader study on restyling demonstrations and is available under the NVIDIA Open Model License.
NVIDIA has released a trained policy model for the Cosmos3-Nano framework, specifically adapted for the LIBERO-10 environment. The model was trained on a combination of real and synthetic data, achieving success rates of 93.9% in standard scenes and 84.7% in perturbed scenes. The model is part of a broader study on restyling demonstrations and is available under the NVIDIA Open Model License.
Source: NVIDIA — Robotics model repository updates · Read original article ↗
Article text · Original source · English
Cosmos3-Nano LIBERO-10 policy, cell D2
A Cosmos3-Nano action policy post-trained on LIBERO-10, one cell of the controlled synthetic-data study in the Cosmos Cookbook recipe "Restyle LIBERO-10 Demonstrations with Cosmos3-Nano-Sim-Transfer and Measure the Gain Against a Matched Control". Export of the training checkpoint (bf16 safetensors, EMA weights, 7 shards, about 30 GB); loadable by the cosmos-framework policy server. The training data is not redistributed here: LIBERO-10 is at nvidia/LIBERO_LeRobot_v3, and the restyled sets are regenerated by the recipe.
| Training data | 379 real LIBERO-10 demonstrations + 1137 duplicates from iteration 1000 |
| Iterations | resume R at 1000 -> 2000 (global batch 2048, training seed 42) |
| Success, standard scenes | 93.9 +/- 0.8 % (10 tasks x 50 initial states, three evaluation seeds) |
| Success, perturbed scenes | 84.7 +/- 2.3 % (same initial states with randomised lighting, textures, skybox and camera jitter) |
| Base model | nvidia/Cosmos3-Nano, revision e59a53c2 |
| Framework | cosmos-framework c23e51f2 (v1.2.2) |
Read the recipe before comparing cells: a restyled cell is only interpretable next to its duplicate control, and differences under 2.5 points are within the training-seed spread.
Licence: derivative of nvidia/Cosmos3-Nano; the NVIDIA Open Model License applies.
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
global batch size
2,048
View original evidence
ver. The training data is not redistributed here: LIBERO-10 is at nvidia/LIBERO_LeRobot_v3 , and the restyled sets are regenerated by the recipe. Training data 379 real LIBERO-10 demonstrations + 1137 duplicates from iteration 1000 Iterations resume R at 1000 -> 2000 (global batch 2048, training seed 42) Success, standard scenes 93.9 +/- 0.8 % (10 tasks x 50 initial states, three evaluation seeds) Success, perturbed
Open source S3training seed
42
View original evidence
ver. The training data is not redistributed here: LIBERO-10 is at nvidia/LIBERO_LeRobot_v3 , and the restyled sets are regenerated by the recipe. Training data 379 real LIBERO-10 demonstrations + 1137 duplicates from iteration 1000 Iterations resume R at 1000 -> 2000 (global batch 2048, training seed 42) Success, standard scenes 93.9 +/- 0.8 % (10 tasks x 50 initial states, three evaluation seeds) Success, perturbed
Open source S3evaluation seeds
3
View original evidence
ver. The training data is not redistributed here: LIBERO-10 is at nvidia/LIBERO_LeRobot_v3 , and the restyled sets are regenerated by the recipe. Training data 379 real LIBERO-10 demonstrations + 1137 duplicates from iteration 1000 Iterations resume R at 1000 -> 2000 (global batch 2048, training seed 42) Success, standard scenes 93.9 +/- 0.8 % (10 tasks x 50 initial states, three evaluation seeds) Success, perturbed
Open source S3number of tasks
10
View original evidence
ver. The training data is not redistributed here: LIBERO-10 is at nvidia/LIBERO_LeRobot_v3 , and the restyled sets are regenerated by the recipe. Training data 379 real LIBERO-10 demonstrations + 1137 duplicates from iteration 1000 Iterations resume R at 1000 -> 2000 (global batch 2048, training seed 42) Success, standard scenes 93.9 +/- 0.8 % (10 tasks x 50 initial states, three evaluation seeds) Success, perturbed
Open source S3
- code: not_reported
- weights: not_reported
- dataset: not_reported
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
ver. The training data is not redistributed here: LIBERO-10 is at nvidia/LIBERO_LeRobot_v3 , and the restyled sets are regenerated by the recipe. Training data 379 real LIBERO-10 demonstrations + 1137 duplicates from iteration 1000 Iterations resume R at 1000 -> 2000 (global batch 2048, training seed 42) Success, standard scenes 93.9 +/- 0.8 % (10 tasks x 50 initial states, three evaluation seeds) Success, perturbed
Open source S3
Source：NVIDIA — Robotics model repository updates · huggingface.co