metadata
license: apache-2.0
datasets:
- TianxingChen/RoboTwin2.0
language:
- en
base_model:
- lerobot/pi05_base
FlashVLA · π0.5 · RoboTwin 2.0
A π0.5 flow-matching vision-language-action policy finetuned on RoboTwin 2.0 (50-task multitask) and served with FlashVLA streaming action decoding for fast, asynchronous inference.
- Base model:
lerobot/pi05_base - Method: FlashVLA — streaming action decoding for flow-matching VLAs (async chunk-overlap execution)
- Benchmark: RoboTwin 2.0, 50-task multitask (clean / randomized)
- License: Apache-2.0
Results
RoboTwin 2.0 50-task multitask success rate (%). d is the async step delay:
d=0 is synchronous, d=1/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 the FlashVLA library.
See sim_eval/robotwin/ for evaluation setup.
Citation
@article{flashvla2026,
title = {{FlashVLA: Streaming Action Decoding for Fast and Asynchronous VLA Inference}},
author = {Li, Zekai and Tang, Jiaming and Liu, Zhijian},
year = {2026}
}
License
Apache-2.0