README.md

---
library_name: lerobot
license: apache-2.0
language:
  - en
base_model:
  - SberRoboticsCenter/Qwen3-VL-4B-Instruct-action
pipeline_tag: robotics
tags:
  - robotics
  - vla
  - vision-language-action
  - manipulation
  - flow-matching
  - action-prediction
  - green-vla
datasets:
  - bridge
  - fractal
---

<div align="center">

# GreenVLA-5b-base-stride-1

### Staged Vision-Language-Action Model for Generalist Robots

**Sber Robotics Center &middot; Manipulation Team**

[![arXiv](https://img.shields.io/badge/arXiv-2602.00919-b31b1b?style=for-the-badge&logo=arxiv&logoColor=white)](https://arxiv.org/abs/2602.00919)
[![Project Page](https://img.shields.io/badge/Project-Page-blue?style=for-the-badge&logo=github&logoColor=white)](https://greenvla.github.io/)
[![Code](https://img.shields.io/badge/Code-GitHub-181717?style=for-the-badge&logo=github&logoColor=white)](https://github.com/greenvla/GreenVLA)

</div>

---

## Overview

**GreenVLA-5b-base-stride-1** is the recommended base checkpoint of the [Green-VLA](https://arxiv.org/abs/2602.00919) family — a ~5B-parameter Vision-Language-Action model pretrained on both general-domain and robotics data (3,000+ hours of demonstrations across multiple embodiments).

This is the **stride-1** variant: the action expert has the **same number of transformer layers** as the VLM backbone, providing maximum action-prediction capacity. For a lighter-weight alternative with 4× fewer action-expert layers, see [GreenVLA-5b-base-stride-4](https://huggingface.co/SberRoboticsCenter/GreenVLA-5b-base-stride-4).

This checkpoint combines:

- **VLM capabilities** — Visual Question Answering, object pointing, bounding box prediction, and scene description, inherited from the [Qwen3-VL-4B](https://huggingface.co/SberRoboticsCenter/Qwen3-VL-4B-Instruct-action) backbone.
- **Autoregressive action prediction** — FAST token-based action generation for discrete control.
- **Flow-matching action expert** — A continuous action head for smooth, high-frequency trajectory generation.

Use this checkpoint as the starting point for **fine-tuning on your own embodiment** (R1 stage), or for zero-shot VLM inference.

## Architecture

| Component | Details |
|---|---|
| **VLM Backbone** | Qwen3-VL-4B-Instruct (vision encoder + language model) |
| **Action Expert** | Flow-matching transformer operating in a reduced hidden space |
| **Action Expert Depth** | Same number of layers as the VLM (stride 1) |
| **Action Tokenizer** | FAST tokenizer for autoregressive action prediction |
| **Total Parameters** | ~5B |

## Training Curriculum

This checkpoint corresponds to the **Base** stage of the Green-VLA curriculum:

| Stage | Name | Status |
|:---:|---|:---:|
| **L0** | Foundational VLM pretraining | ✓ |
| **L1** | Multimodal grounding (VQA, pointing, bbox) | ✓ |
| **R0** | Multi-embodiment robotics pretraining | ✓ |
| R1 | Embodiment-specific adaptation | — |
| R2 | RL policy alignment | — |

## Quick Start

### Installation

```bash
git clone https://github.com/greenvla/GreenVLA.git
cd GreenVLA
uv sync  # or: pip install -e .
```

### Action Inference

```python
import numpy as np
import torch
from lerobot.common.policies.factory import load_pretrained_policy
from lerobot.common.utils.torch_observation import (
    move_dict_to_batch_for_inference,
    torch_preprocess_dict_inference,
)

# 1. Load policy and transforms.
policy, input_transforms, output_transforms = load_pretrained_policy(
    "SberRoboticsCenter/GreenVLA-5b-stride-1-R1-bridge",
    data_config_name="bridge",
)
policy.to("cuda").eval()

# 2. Build an observation (replace with real sensor data).
raw_obs = {
    "observation/state": np.random.rand(8).astype(np.float32),  # x y z roll pitch yaw _pad_ gripper
    "observation/image": np.random.randint(0, 256, size=(224, 224, 3), dtype=np.uint8),
    "prompt": "pick up the green block and place it on the plate",
}

# 3. Transform, preprocess, and batch.
obs = input_transforms(raw_obs)
obs = torch_preprocess_dict_inference(obs)
batch = move_dict_to_batch_for_inference(obs, device="cuda")

# 4. Predict actions and post-process.
with torch.inference_mode():
    raw_actions = policy.select_action(batch).cpu().numpy()

actions = output_transforms(
    {"actions": raw_actions, "state": batch["state"].cpu().numpy()}
)["actions"]
# actions shape: (action_horizon, 7) — [x, y, z, roll, pitch, yaw, gripper]
```

See [`examples/example_inference_bridge.py`](https://github.com/greenvla/GreenVLA/blob/main/examples/example_inference_bridge.py) for the full runnable script with argument parsing.

### VLM Inference (VQA, Pointing, BBox)

The base model retains full VLM capabilities:

```python
from PIL import Image
from lerobot.common.policies.factory import load_pretrained_policy

# Load without data transforms
policy, _, _ = load_pretrained_policy(
    "SberRoboticsCenter/GreenVLA-5b-base-stride-1",
    data_config_name=None,
)
policy = policy.to("cuda").eval()

# Access the processor and model directly
processor = policy.model.processor
image = Image.open("scene.jpg")

messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "image": image},
            {"type": "text", "text": "Describe what the robot should do next."},
        ],
    }
]

inputs = processor.apply_chat_template(
    messages, tokenize=True, add_generation_prompt=False,
    return_dict=True, return_tensors="pt",
    padding_side="left", padding="max_length", max_length=256,
    images_kwargs={"do_resize": True},
).to("cuda")

generated_ids = policy.model.model.generate(
    **inputs, max_new_tokens=256, do_sample=False, use_cache=False,
)

generated_ids_trimmed = [
    out[len(inp):] for inp, out in zip(inputs.input_ids, generated_ids)
]
print(processor.batch_decode(generated_ids_trimmed, skip_special_tokens=True)[0])
```

## Citation

```bibtex
@misc{apanasevich2026greenvlastagedvisionlanguageactionmodel,
    title   = {Green-VLA: Staged Vision-Language-Action Model for Generalist Robots},
    author  = {I. Apanasevich and M. Artemyev and R. Babakyan and P. Fedotova and
               D. Grankin and E. Kupryashin and A. Misailidi and D. Nerus and
               A. Nutalapati and G. Sidorov and I. Efremov and M. Gerasyov and
               D. Pikurov and Y. Senchenko and S. Davidenko and D. Kulikov and
               M. Sultankin and K. Askarbek and O. Shamanin and D. Statovoy and
               E. Zalyaev and I. Zorin and A. Letkin and E. Rusakov and
               A. Silchenko and V. Vorobyov and S. Sobolnikov and A. Postnikov},
    year    = {2026},
    eprint  = {2602.00919},
    archivePrefix = {arXiv},
    primaryClass  = {cs.RO},
    url     = {https://arxiv.org/abs/2602.00919},
}
```

<div align="center">

&copy; 2026 Sber Robotics Center &middot; Manipulation Team

</div>