Views
No views yet
🎉 Accepted to CVPR 2026. ✍️ Fei Ni¹, Zhuo Chen², Yifu Yuan³, Zibin Dong³, Xianze Yao³, Shan Luo², Jianye Hao³, Jiankang Deng¹†, Stefanos Zafeiriou¹†
🏫 ¹Imperial College London ²King's College London ³Tianjin University
✉️ Primary contact: f.ni@imperial.ac.uk
| Suite | Success rate |
|---|---|
libero_spatial | 0.988 |
libero_object | 0.996 |
libero_goal | 0.974 |
libero_10 | 0.970 |
| 4-suite mean | 0.982 |
| Component | Choice |
|---|---|
| VLM backbone | Qwen3-VL-4B-Instruct |
| Action head | DiT-B (flow matching) |
| LAM tokenizer | 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 | 8 |
SemanticVLA-LIBERO/
├── README.md
├── config.yaml # loadable model config
├── dataset_statistics.json # action normalization stats
└── final_model/
└── pytorch_model.pt # policy state_dict1from semanticvla.model.framework.base_framework import baseframework
2
3policy = baseframework.from_pretrained("pytorch_model.pt")
4policy.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.examples/LIBERO/ in the code repo.| Repo | Purpose |
|---|---|
🤗 SemanticVLA-LAM | Unified OXE LAM consumed by this policy |
🤗 SemanticVLA-SimplerEnv | SimplerEnv WidowX policy |
1@inproceedings{ni2026semanticvla,
2 title = {SemanticVLA: Towards Semantic Reasoning over Action Memorization via Synergistic Explicit Trace and Latent Action Planning},
3 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},
4 booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
5 year = {2026}
6}