Downloads · 30 days
0
grab_roller-cross-pretrain-42
Zhaohf03/grab_roller-cross-pretrain-42
grab_roller-cross-pretrain-42 is a machine learning model from Zhaohf03. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
Cross-embodiment diffusion policy checkpoints for grabroller (4 embodiments: aloha-agilex, franka, piper, arx-x5).
Access
Public
Updated Jun 5, 2026
Repo size
2.9 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.ckpt2.9 GB · 100%
At a glance
- Access
- Public
- Created
- Jun 5, 2026
- Updated
- Jun 5, 2026
- SHA
- 6034f73d
Model card
From the Hugging Face model README
grab_roller LBM cross-embodiment pretrain (seed=42)
Cross-embodiment diffusion policy checkpoints for grab_roller (4 embodiments: aloha-agilex, franka, piper, arx-x5).
Training config
| Item | Value |
|---|---|
| Config | grab_roller_cross_pretrain.yaml |
| Vision | DINOv2 vision-only, 3 cameras (head + dual wrist) |
| Action | chunkwise_delta_endpose (20d 6drot) |
| Proprio | ee, relative |
| DDP | 8× GPU, batch 48/rank (global 384) |
| LR | 8e-4 |
| Seed | 42 |
Checkpoints
| File | Epoch |
|---|---|
200.ckpt | 200 |
300.ckpt | 300 |
400.ckpt | 400 |
500.ckpt | 500 |
Deploy (XRoboTwin)
cd policy/LBM
python link_ckpt.py \
--task grab_roller --ckpt-setting cross4 \
--expert-data-num 50 --seed 42 --epoch 400 \
--hydra-ckpt checkpoints/grab_roller-cross-42/400.ckpt
Eval with --embodiment_name arx-x5 (or other cross4 embodiment) for cross-body normalizer.
Repo: figsama/XRoboTwin branch figsama.
- Created
- Jun 5, 2026
- Updated
- Jun 5, 2026