Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,51 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
base_model: HuggingFaceTB/SmolVLM-Instruct
|
| 4 |
+
tags:
|
| 5 |
+
- lerobot
|
| 6 |
+
- smolvla
|
| 7 |
+
- vla
|
| 8 |
+
- robotics
|
| 9 |
+
- manipulation
|
| 10 |
+
- real-robot
|
| 11 |
+
- so101
|
| 12 |
+
- visuomotor
|
| 13 |
+
- language-conditioned
|
| 14 |
+
datasets:
|
| 15 |
+
- ShubhamK32/so101_declutter_v1
|
| 16 |
+
pipeline_tag: robotics
|
| 17 |
+
---
|
| 18 |
+
|
| 19 |
+
# SmolVLA — SO-101 Space Decluttering
|
| 20 |
+
|
| 21 |
+
SmolVLA policy fine-tuned on the [SO-101 Space Decluttering Dataset v1](https://huggingface.co/datasets/ShubhamK32/so101_declutter_v1) for language-conditioned pick-and-place decluttering tasks on a 6-DoF SO-101 robotic arm. Trained using [LeRobot](https://github.com/huggingface/lerobot).
|
| 22 |
+
|
| 23 |
+
## Training Details
|
| 24 |
+
|
| 25 |
+
- **Policy:** SmolVLA (Vision-Language-Action)
|
| 26 |
+
- **Steps:** 20,000
|
| 27 |
+
- **Robot:** SO-101 6-DoF leader-follower
|
| 28 |
+
- **Cameras:** Dual-view — fixed top-view + wrist-mounted egocentric
|
| 29 |
+
- **Framework:** LeRobot
|
| 30 |
+
- **Language conditioning:** Task descriptions passed as natural language instructions
|
| 31 |
+
|
| 32 |
+
## Dataset
|
| 33 |
+
|
| 34 |
+
Trained on [ShubhamK32/so101_declutter_v1](https://huggingface.co/datasets/ShubhamK32/so101_declutter_v1) — a multi-view teleoperation dataset with spatial distractors injected to prevent visual shortcut learning.
|
| 35 |
+
|
| 36 |
+
## Usage
|
| 37 |
+
```python
|
| 38 |
+
from lerobot.policies.smolvla.modeling_smolvla import SmolVLAPolicy
|
| 39 |
+
|
| 40 |
+
policy = SmolVLAPolicy.from_pretrained("ShubhamK32/smolvla_so101_declutter")
|
| 41 |
+
```
|
| 42 |
+
|
| 43 |
+
## Camera Views
|
| 44 |
+
|
| 45 |
+
- `observation.images.topview` — Fixed overhead. Better for unoccluded pick-place tasks.
|
| 46 |
+
- `observation.images.wristview` — Egocentric wrist-mounted. Better for overlapping and cluttered scenes.
|
| 47 |
+
|
| 48 |
+
## Related
|
| 49 |
+
|
| 50 |
+
- Dataset: [ShubhamK32/so101_declutter_v1](https://huggingface.co/datasets/ShubhamK32/so101_declutter_v1)
|
| 51 |
+
- ACT checkpoint: [ShubhamK32/act_so101_declutter](https://huggingface.co/ShubhamK32/act_so101_declutter)
|