Open-HumanWAM
Collection
Model checkpoints for Open-HumanWAM. • 5 items • Updated • 2
HumanWAM-Vanilla-Base
HumanWAM-Vanilla checkpoints pretrained on 50,000 hours of Open-HumanScale-50K, followed by EgoPro multi-view adaptation. These checkpoints provide the human-experience initialization for downstream robot post-training.
Checkpoints
| File | Stage | Selected step | Initialization |
|---|---|---|---|
pretrained/step_040000.pt |
Human egocentric pretraining | 40,000 | Human pretraining stage |
midtrained/step_010000.pt |
EgoPro multi-view mid-training, W0 setting | 10,000 | The pretrained checkpoint above |
Both files are PyTorch .pt checkpoints. Step numbers are local to each stage.
The mid-trained checkpoint initializes the released
HumanWAM-Vanilla-RoboTwin policy.
Usage
hf download DAGroup-PKU/HumanWAM-Vanilla-Base --local-dir HumanWAM-Vanilla-Base
Use the matching HumanWAM-Vanilla implementation for downstream adaptation. The checkpoints are training initializations; task-specific robot inference requires the corresponding post-trained policy and its normalization statistics, action transforms, and runtime resources. See the code repository for setup and training documentation.
Release status
This is a weights-first release: the pretrained and mid-trained weights are available. Matching normalization statistics, configurations, and runtime resources will be added separately. The downloaded files alone are not a complete runnable package.
File integrity
| File | Bytes | SHA-256 |
|---|---|---|
pretrained/step_040000.pt |
12041838137 | e0a00e55483d8c08376be2650f159da1d2ce4ec22d5cfcf925a28a310de6cbf6 |
midtrained/step_010000.pt |
12041838009 | c87cee0ee7fac06ed55f809230e2c83c26f69b4c3239ce561e191df8aaccaf9f |