---
license: mit
library_name: pytorch
pipeline_tag: robotics
tags:
  - robotics
  - vision-language-action
  - vla
  - simpler-env
  - widowx
  - bridge
  - manipulation
  - qwen-vl
---

# SemanticVLA · SimplerEnv (WidowX)

> 🎉 **Accepted to [CVPR 2026](https://cvpr.thecvf.com/virtual/2026/poster/39352).**
> ✍️ Fei Ni¹, Zhuo Chen², Yifu Yuan³, Zibin Dong³, Xianze Yao³, Shan Luo², Jianye Hao³, Jiankang Deng¹†, Stefanos Zafeiriou¹†<br>
> 🏫 ¹Imperial College London &nbsp;&nbsp; ²King's College London &nbsp;&nbsp; ³Tianjin University<br>
> ✉️ Primary contact: [f.ni@imperial.ac.uk](mailto:f.ni@imperial.ac.uk)

[SemanticVLA](https://github.com/Fei-Ni/SemanticVLA_Offcial) policy trained on BridgeData V2 (Open X-Embodiment `bridge_orig`) for **100K steps**, intended for [SimplerEnv](https://github.com/simpler-env/SimplerEnv) WidowX evaluation. The unified OXE LAM is used as the latent-action tokenizer, and the trace + latent-action auxiliary heads are supervised in the VLM's language stream.

## Headline result (SimplerEnv WidowX)

| Task | Success rate |
|---|---:|
| Put Eggplant in Basket | 0.958 |
| Spoon on Towel         | 1.000 |
| Carrot on Plate        | 0.792 |
| Stack Cube             | 0.458 |
| **Mean**               | **0.802** |

## Architecture

| Component | Choice |
|---|---|
| VLM backbone | Qwen3-VL-4B-Instruct |
| Action head | DiT-B (flow matching) |
| LAM tokenizer | [`SemanticVLA-LAM`](https://huggingface.co/spikefly/SemanticVLA-LAM) (unified OXE LAM) |
| Semantic supervision | Trace + latent action tokens predicted in the VLM's language stream; action decoder unmodified |
| Latent vocabulary size | 32 |
| Latent tokens per sample | 4 |
| Action horizon | 16 |

## Training data

This checkpoint is trained on **BridgeData V2** (Open X-Embodiment `bridge_orig`) for 100K steps. It is intended specifically for SimplerEnv WidowX evaluation and is **not** meant as a general-purpose policy for unrelated robot embodiments.

## Files

```
SemanticVLA-SimplerEnv/
├── README.md
├── config.yaml              # loadable model config
├── dataset_statistics.json  # action normalization stats
└── final_model/
    └── pytorch_model.pt     # policy state_dict
```

## How to load

```python
from semanticvla.model.framework.base_framework import baseframework

policy = baseframework.from_pretrained("pytorch_model.pt")
policy.eval()
```

`baseframework.from_pretrained()` walks two directory levels up from the checkpoint file to locate `config.yaml` and `dataset_statistics.json`. The released layout follows this convention.

To run the SimplerEnv WidowX suite, see [`examples/SimplerEnv/`](https://github.com/Fei-Ni/SemanticVLA_Offcial/tree/main/examples/SimplerEnv) in the code repo.

## Sibling SemanticVLA checkpoint repos

| Repo | Purpose |
|---|---|
| 🤗 [`SemanticVLA-LAM`](https://huggingface.co/spikefly/SemanticVLA-LAM) | Unified OXE LAM consumed by this policy |
| 🤗 [`SemanticVLA-LIBERO`](https://huggingface.co/spikefly/SemanticVLA-LIBERO) | LIBERO policy |

## Related resources

- **Code**: https://github.com/Fei-Ni/SemanticVLA_Offcial
- **Dataset (BridgeData V2 in LeRobot v3 with dense traces)**: 🤗 [`SemanticVLA-TraceX-240K-Bridge`](https://huggingface.co/datasets/spikefly/SemanticVLA-TraceX-240K-Bridge)
- **Datasets collection**: https://hf.co/collections/spikefly/semanticvla-datasets
- **Model Zoo collection**: https://hf.co/collections/spikefly/semanticvla-model-zoo

## Citation

```bibtex
@inproceedings{ni2026semanticvla,
  title     = {SemanticVLA: Towards Semantic Reasoning over Action Memorization via Synergistic Explicit Trace and Latent Action Planning},
  author    = {Ni, Fei and Chen, Zhuo and Yuan, Yifu and Dong, Zibin and Yao, Xianze and Luo, Shan and Hao, Jianye and Deng, Jiankang and Zafeiriou, Stefanos},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year      = {2026}
}
```

## License

Released under the [MIT License](https://github.com/Fei-Ni/SemanticVLA_Offcial/blob/main/LICENSE), subject to the upstream BridgeData V2 license.