metadata
license: apache-2.0
task_categories:
- robotics
tags:
- vla
- vision-language-action
- lerobot
- manipulation
- smolvla
size_categories:
- n<1K
SmolVLA Base
SmolVLA — A compact Vision-Language-Action model from Hugging Face for robot manipulation. Takes images + language instructions, outputs robot actions.
Model Details
| Field | Value |
|---|---|
| License | Apache 2.0 (commercial use ✅) |
| Input | Multi-view images (256×256) + language instruction + proprioception |
| Output | Robot actions |
| Framework | LeRobot / PyTorch |
| Format | Safetensors |
Quick Start
from huggingface_hub import snapshot_download
from lerobot.policies.smolvla.modeling_smolvla import SmolVLA
# Download model
snapshot_download(repo_id="BlackCatRoboticsAI/smolvla-base", local_dir="./smolvla")
# Load policy
policy = SmolVLA.from_pretrained("./smolvla")
Use Cases
- Pick-and-place manipulation
- Object sorting
- Assembly tasks
- Language-driven control
Citation
@misc{lerobot2025smolvla,
title={SmolVLA: Small Vision-Language-Action Model},
author={Hugging Face},
year={2025},
publisher={Hugging Face}
}