---
license: gemma
pipeline_tag: robotics
language:
- en
inference: false
base_model:
- lerobot/pi05_base
datasets:
- TianxingChen/RoboTwin2.0
tags:
- flashvla
- vision-language-action
- robotics
- flow-matching
- streaming-action-decoding
- pi05
---

# FlashVLA · π0.5 · RoboTwin 2.0

**Streaming Action Decoding for Fast and Asynchronous VLA Inference**

[![Paper](https://img.shields.io/badge/arXiv-2608.27384-b31b1b.svg)](https://arxiv.org/abs/2608.27384)
[![GitHub](https://img.shields.io/badge/GitHub-FlashVLA-181717?logo=github)](https://github.com/z-lab/flashvla)
[![Blog](https://img.shields.io/badge/Blog-FlashVLA-blue)](https://z-lab.ai/projects/flashvla/)
[![Models](https://img.shields.io/badge/%F0%9F%A4%97-Models-yellow)](https://huggingface.co/collections/z-lab/flashvla)

A **π0.5** flow-matching vision-language-action policy finetuned on **RoboTwin 2.0** (50-task multitask) and
served with [FlashVLA](https://github.com/z-lab/flashvla) streaming action decoding for fast, asynchronous inference.

- **Base model:** [`lerobot/pi05_base`](https://huggingface.co/lerobot/pi05_base)
- **Method:** [FlashVLA](https://github.com/z-lab/flashvla), streaming action decoding for flow-matching VLAs (async chunk-overlap execution)
- **Benchmark:** RoboTwin 2.0, 50-task multitask (clean / randomized)

## Results

RoboTwin 2.0 50-task multitask success rate (%). `d` is the async step delay:
`d=0` is synchronous, `d=1` and `d=2` overlap the next chunk's inference with execution.

| Model | Clean | Random | Avg |
|:--|:--:|:--:|:--:|
| π0.5 (base) | 82.74 | 76.76 | 79.75 |
| **+FlashVLA** (`d=0`) | 90.64 | 90.06 | 90.35 |
| **+FlashVLA** (`d=1`) | **91.14** | **90.60** | **90.87** |
| **+FlashVLA** (`d=2`) | 90.20 | 89.66 | 89.93 |

## Usage

Install FlashVLA:

```bash
git clone https://github.com/z-lab/flashvla.git
cd flashvla
conda env create -f environment.yml
conda activate flashvla
```

RoboTwin 2.0 evaluation uses a server/client split across two environments. After the one-time setup in
[`sim_eval/robotwin/`](https://github.com/z-lab/flashvla/tree/main/sim_eval/robotwin), from
`$ROBOTWIN/policy/pi05_flashvla/`:

```bash
bash eval_server.sh                      # terminal 1, flashvla env, starts the policy server
ROBOTWIN_VENV=... bash eval_client.sh    # terminal 2, RoboTwin env, runs the SAPIEN sim
```

Training configs for this checkpoint are in
[`train/configs/pi05/robotwin/`](https://github.com/z-lab/flashvla/tree/main/train/configs/pi05/robotwin).

## License

These weights are finetuned from [`lerobot/pi05_base`](https://huggingface.co/lerobot/pi05_base), which is
released under the [Gemma Terms of Use](https://ai.google.dev/gemma/terms). Those terms govern model
derivatives, so they apply to this checkpoint and to anything derived from it, including the
[Gemma Prohibited Use Policy](https://ai.google.dev/gemma/prohibited_use_policy). If you redistribute this
checkpoint or a derivative of it, you must pass the same terms along.

The [FlashVLA](https://github.com/z-lab/flashvla) inference and training code is separately released under the
[Apache 2.0 License](https://github.com/z-lab/flashvla/blob/main/LICENSE).

## Citation

```bibtex
@inproceedings{li2026flashvla,
  title     = {{FlashVLA: Streaming Action Decoding for Fast and Asynchronous VLA Inference}},
  author    = {Li, Zekai and Tang, Jiaming and Liu, Zhijian},
  booktitle = {Conference on Robot Learning (CoRL)},
  year      = {2026}
}
```