NVIDIA Releases Cosmos3-Nano Policy Model for LIBERO-10 Sim-to-Real Transfer
nvidia/cosmos3-nano-libero10-sdg-export-D_half
NVIDIA has released a policy model trained on the LIBERO-10 environment, part of the Cosmos Cookbook recipe, with performance metrics across standard and perturbed scenes. The model is available under the NVIDIA Open Model License and can be loaded by the cosmos-framework policy server.
NVIDIA provides an open-source policy model trained on the LIBERO-10 environment, with performance metrics reported across standard and perturbed scenes, offering a benchmark for sim-to-real transfer in robotics learning.
Source: NVIDIA — Robotics model repository updates · Read original article ↗
Article text · Original source · English
Cosmos3-Nano LIBERO-10 policy, cell D_half
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 + 568 duplicates |
| Iterations | 0->2000 (global batch 2048, training seed 42) |
| Success, standard scenes | 94.0 +/- 0.5 % (10 tasks x 50 initial states, three evaluation seeds) |
| Success, perturbed scenes | 85.9 +/- 2.6 % (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
training iterations
>2,000
View original evidence
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 + 568 duplicates Iterations 0->2000 (global batch 2048, training seed 42) Success, standard scenes 94.0 +/- 0.5 % (10 tasks x 50 initial states, three evaluation seeds) Success, perturbed scenes 85.9 +/- 2.6 % (same initia
Open source S3global batch size
2,048
View original evidence
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 + 568 duplicates Iterations 0->2000 (global batch 2048, training seed 42) Success, standard scenes 94.0 +/- 0.5 % (10 tasks x 50 initial states, three evaluation seeds) Success, perturbed scenes 85.9 +/- 2.6 % (same initia
Open source S3training seed
42
View original evidence
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 + 568 duplicates Iterations 0->2000 (global batch 2048, training seed 42) Success, standard scenes 94.0 +/- 0.5 % (10 tasks x 50 initial states, three evaluation seeds) Success, perturbed scenes 85.9 +/- 2.6 % (same initia
Open source S3evaluation seeds
3
View original evidence
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 + 568 duplicates Iterations 0->2000 (global batch 2048, training seed 42) Success, standard scenes 94.0 +/- 0.5 % (10 tasks x 50 initial states, three evaluation seeds) Success, perturbed scenes 85.9 +/- 2.6 % (same initia
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.
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 + 568 duplicates Iterations 0->2000 (global batch 2048, training seed 42) Success, standard scenes 94.0 +/- 0.5 % (10 tasks x 50 initial states, three evaluation seeds) Success, perturbed scenes 85.9 +/- 2.6 % (same initia
Open source S3
Source：NVIDIA — Robotics model repository updates · huggingface.co