---
title: "Stable Language Guidance for Vision–Language–Action Models"
description: "They propose Residual Semantic Steering (RSS), which has two parts: Monte Carlo Syntactic Integration (MCSI) , a training… 39 references."
authors: ["Zhihao Zhan", "Yuhao Chen", "Jiaying Zhou", "Qinhan Lyu", "Hao Liu", "Keze Wang", "Liang Lin", "Guangrun Wang"]
year: 2026
venue: "Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)"
source: "https://aclanthology.org/2026.acl-long.190/"
doi: "10.18653/v1/2026.acl-long.190"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/zhan-2026-stable-language-guidance-vision-language/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/zhan-2026-stable-language-guidance-vision-language/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/50/) / Stable Language Guidance for Vision–Language–Action Models

# Stable Language Guidance for Vision–Language–Action Models

**Authors:** Zhihao Zhan, Yuhao Chen, Jiaying Zhou, Qinhan Lyu, [Hao Liu](https://md.research.elamur.ai/free/en/authors/hao-liu/), [Keze Wang](https://md.research.elamur.ai/free/en/authors/keze-wang/), [Liang Lin](https://md.research.elamur.ai/free/en/authors/liang-lin/) and Guangrun Wang · **Year:** 2026 · **Venue:** [Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)](https://md.research.elamur.ai/free/en/venues/acl/) ([ACL 2026](https://md.research.elamur.ai/free/en/venues/acl/2026/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2026.acl-long.190/) · [PDF](https://aclanthology.org/2026.acl-long.190.pdf) · [DOI 10.18653/v1/2026.acl-long.190](https://doi.org/10.18653/v1/2026.acl-long.190) · [OpenAlex](https://openalex.org/W7166816475)

**Citations:** 0 citations (OpenAlex, 2026-10-14)

## Summary
Vision–Language–Action (VLA) robot policies often fail when an instruction is reworded, partly masked, or points to an unfamiliar goal. The authors attribute this to two causes. Training data covers only a narrow slice of possible phrasings, and dense visual input overpowers the sparse language signal. As a result, the policy tends to act on what the scene affords rather than on what was asked. They propose Residual Semantic Steering (RSS), which has two parts:

- **Monte Carlo Syntactic Integration (MCSI)**, a training step that uses an LLM/VLM teacher to expand each instruction into many paraphrases, layout descriptions and subtask breakdowns. The loss is then averaged over this neighbourhood.
- **Residual Affordance Steering (RAS)**, an inference step that subtracts the action scores from a pass with blank language input. This keeps only the shift that the instruction itself causes.

Built on π0.5 and evaluated on LIBERO, RSS loses far less performance under paraphrased instructions than the baseline. This makes language-conditioned robot control more dependable.

## Key points
- The paper groups instruction perturbations into three failure types: destructive overwriting (masked tokens), obfuscated reinterpretation (synonyms or verbose wording), and out-of-distribution semantic transfer (unseen goals).
- Paraphrases were generated by models not used in training, with no retraining:
  - With DeepSeek-R1 paraphrases, baseline π0.5 dropped 10.54%, while the +MCSI model dropped only 1.28%.
  - With Qwen3.5 paraphrases, the baseline dropped 5.10% and +MCSI dropped 1.71%. The RAS & MCSI model gained 1.41%, reaching a 75.00% average success rate.
- Results barely changed when the teacher used to generate neighbourhoods was swapped. Using InternVL3 or LLaVA-OneVision instead of Qwen2.5-VL shifted performance by +2.43% and −0.54% respectively (average success rate 73.55–75.75%).
- A stated limitation is that RAS can hesitate or stay idle on vague prompts such as "do something". The authors present this as safer than defaulting to habitual visual actions.
- The code is available at github.com/Doo-mon/RSS.

*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*

## Institutions

- [Sun Yat-sen University](https://md.research.elamur.ai/free/en/universities/sun-yat-sen-university/) — University, Guangzhou, China · 72 works · 9,502 citations

📚 Shelves (5): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 18,614 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 17,965 works · [👁️ Computer Vision and Pattern Recognition](https://md.research.elamur.ai/free/en/shelves/computer-science/computer-vision-and-pattern-recognition/) · 3,486 works · [⚙️ Engineering](https://md.research.elamur.ai/free/en/shelves/engineering/) · 188 works · [⚙️ Control and Systems Engineering](https://md.research.elamur.ai/free/en/shelves/engineering/control-and-systems-engineering/) · 62 works

🧭 Topics (1): [🖼️ Vision & language](https://md.research.elamur.ai/free/en/topics/multimodal/) · 811 works

## Cite

Cite as: Zhihao Zhan, Yuhao Chen, Jiaying Zhou, Qinhan Lyu, Hao Liu, Keze Wang, Liang Lin and Guangrun Wang. 2026. Stable Language Guidance for Vision–Language–Action Models. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{zhan2026stable,
  title = {{Stable Language Guidance for Vision–Language–Action Models}},
  author = {Zhihao Zhan, Yuhao Chen, Jiaying Zhou, Qinhan Lyu, Hao Liu, Keze Wang, Liang Lin and Guangrun Wang},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-long.190},
  url = {https://doi.org/10.18653/v1/2026.acl-long.190}
}
```

- [← previous Stabilizing Efficient Reasoning with Step-Level Advantage Selection](https://md.research.elamur.ai/free/en/papers/wang-2026-stabilizing-efficient-reasoning-step-level/)
- [next → STAGEET: Stage-wise Typed Edit Tagging for Grammatical Error Correction with Arabic as a Case Study](https://md.research.elamur.ai/free/en/papers/lou-2026-stageet-stage-wise-typed-edit/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/50/)

## References

39 references, 7 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, and 1 others. 2025. Qwen2. 5-vl technical report. arXiv preprint arXiv:2502.13923. — [Qwen2.5-VL Technical Report](https://md.research.elamur.ai/free/en/papers/bai-2025-qwen2-5-vl-technical/)
2. Johan Bjorck, Fernando Castañeda, Nikita Cherniadev, Xingye Da, Runyu Ding, Linxi Fan, Yu Fang, Dieter Fox, Fengyuan Hu, Spencer Huang, and 1 others. 2025. Gr00t n1: An open foundation model for generalist humanoid robots. arXiv preprint arXiv:2503.14734. [find](https://scholar.google.com/scholar?q=Johan+Bjorck%2C+Fernando+Casta%C3%B1eda%2C+Nikita+Cherniadev%2C+Xingye+Da%2C+Runyu+Ding%2C+Linxi+Fan%2C+Yu+Fang%2C+Dieter+Fox%2C+Fengyuan+Hu%2C+Spencer+Huang%2C+and+1+others%2E+2025%2E+Gr00t+n1%3A+An+open+foundation+model+for+gener)
3. Kevin Black, Noah Brown, Danny Driess, Adnan Esmail, Michael Equi, Chelsea Finn, Niccolo Fusai, Lachy Groom, Karol Hausman, Brian Ichter, and 1 others. 2024. Pi0: A Vision-Language-Action Flow Model for General Robot Control. arXiv preprint arXiv:2410.24164. [find](https://scholar.google.com/scholar?q=Kevin+Black%2C+Noah+Brown%2C+Danny+Driess%2C+Adnan+Esmail%2C+Michael+Equi%2C+Chelsea+Finn%2C+Niccolo+Fusai%2C+Lachy+Groom%2C+Karol+Hausman%2C+Brian+Ichter%2C+and+1+others%2E+2024%2E+Pi0%3A+A+Vision%2DLanguage%2DAction+Flow+Model+f)
4. Anthony Brohan, Noah Brown, Justice Carbajal, Yevgen Chebotar, Joseph Dabis, Chelsea Finn, Keerthana Gopalakrishnan, Karol Hausman, Alex Herzog, Jasmine Hsu, and 1 others. 2022. Rt-1: Robotics transformer for real-world control at scale. arXiv preprint arXiv:2212.06817. [find](https://scholar.google.com/scholar?q=Anthony+Brohan%2C+Noah+Brown%2C+Justice+Carbajal%2C+Yevgen+Chebotar%2C+Joseph+Dabis%2C+Chelsea+Finn%2C+Keerthana+Gopalakrishnan%2C+Karol+Hausman%2C+Alex+Herzog%2C+Jasmine+Hsu%2C+and+1+others%2E+2022%2E+Rt%2D1%3A+Robotics+transfo)
5. Qingwen Bu, Yanting Yang, Jisong Cai, Shenyuan Gao, Guanghui Ren, Maoqing Yao, Ping Luo, and Hongyang Li. 2025. Univla: Learning to act anywhere with task-centric latent actions. arXiv preprint arXiv:2505.06111. — [arXiv:2505.06111](https://arxiv.org/abs/2505.06111) [find](https://scholar.google.com/scholar?q=Qingwen+Bu%2C+Yanting+Yang%2C+Jisong+Cai%2C+Shenyuan+Gao%2C+Guanghui+Ren%2C+Maoqing+Yao%2C+Ping+Luo%2C+and+Hongyang+Li%2E+2025%2E+Univla%3A+Learning+to+act+anywhere+with+task%2Dcentric+latent+actions%2E+arXiv+preprint+arXiv%3A)
6. Jun Cen, Chaohui Yu, Hangjie Yuan, Yuming Jiang, Siteng Huang, Jiayan Guo, Xin Li, Yibing Song, Hao Luo, Fan Wang, and 1 others. 2025. Worldvla: Towards autoregressive action world model. arXiv preprint arXiv:2506.21539. — [arXiv:2506.21539](https://arxiv.org/abs/2506.21539) [find](https://scholar.google.com/scholar?q=Jun+Cen%2C+Chaohui+Yu%2C+Hangjie+Yuan%2C+Yuming+Jiang%2C+Siteng+Huang%2C+Jiayan+Guo%2C+Xin+Li%2C+Yibing+Song%2C+Hao+Luo%2C+Fan+Wang%2C+and+1+others%2E+2025%2E+Worldvla%3A+Towards+autoregressive+action+world+model%2E+arXiv+prepri)
7. Yuhao Chen, Zhihao Zhan, Xiaoxin Lin, Zijian Song, Hao Liu, Qinhan Lyu, Yubo Zu, Xiao Chen, Zhiyuan Liu, Tao Pu, and 1 others. 2026. Radar: Benchmarking vision-language-action generalization via real-world dynamics, spatial-physical intelligence, and autonomous evaluation. arXiv preprint arXiv:2602.… — [arXiv:2602.10980](https://arxiv.org/abs/2602.10980) [find](https://scholar.google.com/scholar?q=Yuhao+Chen%2C+Zhihao+Zhan%2C+Xiaoxin+Lin%2C+Zijian+Song%2C+Hao+Liu%2C+Qinhan+Lyu%2C+Yubo+Zu%2C+Xiao+Chen%2C+Zhiyuan+Liu%2C+Tao+Pu%2C+and+1+others%2E+2026%2E+Radar%3A+Benchmarking+vision%2Dlanguage%2Daction+generalization+via+real%2D)
8. Zhe Chen, Jiannan Wu, Wenhai Wang, Weijie Su, Guo Chen, Sen Xing, Muyan Zhong, Qinglong Zhang, Xizhou Zhu, Lewei Lu, and 1 others. 2024. [Internvl: Scaling up vision foundation models and aligning for generic visual-linguistic tasks](https://md.research.elamur.ai/free/en/papers/chen-2023-internvl-scaling-up-vision/). In Proceedings of the IEEE/CVF Conference on Computer Vision and Pa…
9. Cheng Chi, Zhenjia Xu, Siyuan Feng, Eric Cousineau, Yilun Du, Benjamin Burchfiel, Russ Tedrake, and Shuran Song. 2023. [Diffusion policy: Visuomotor policy learning via action diffusion](https://md.research.elamur.ai/free/en/papers/chi-2023-diffusion-policy-visuomotor-policy-learning/). The International Journal of Robotics Research, page 02783649241273668.
10. Senyu Fei, Siyin Wang, Junhao Shi, Zihao Dai, Jikun Cai, Pengfang Qian, Li Ji, Xinzhe He, Shiduo Zhang, Zhaoye Fei, and 1 others. 2025. Libero-plus: In-depth robustness analysis of vision-language-action models. arXiv preprint arXiv:2510.13626. — [arXiv:2510.13626](https://arxiv.org/abs/2510.13626) [find](https://scholar.google.com/scholar?q=Senyu+Fei%2C+Siyin+Wang%2C+Junhao+Shi%2C+Zihao+Dai%2C+Jikun+Cai%2C+Pengfang+Qian%2C+Li+Ji%2C+Xinzhe+He%2C+Shiduo+Zhang%2C+Zhaoye+Fei%2C+and+1+others%2E+2025%2E+Libero%2Dplus%3A+In%2Ddepth+robustness+analysis+of+vision%2Dlanguage%2Dact)
11. Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Peiyi Wang, Qihao Zhu, Runxin Xu, Ruoyu Zhang, Shirong Ma, Xiao Bi, and 1 others. 2025. [Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning](https://md.research.elamur.ai/free/en/papers/deepseek-ai-2025-deepseek-r1-incentivizing-reasoning-capability/). arXiv preprint arXiv:2501.12948.
12. Jonathan Ho and Tim Salimans. 2022. [Classifier-free diffusion guidance](https://md.research.elamur.ai/free/en/papers/ho-2022-classifier-free-diffusion-guidance/). arXiv preprint arXiv:2207.12598.
13. Zhi Hou, Tianyi Zhang, Yuwen Xiong, Haonan Duan, Hengjun Pu, Ronglei Tong, Chengyang Zhao, Xizhou Zhu, Yu Qiao, Jifeng Dai, and 1 others. 2025. Dita: Scaling diffusion transformer for generalist vision-language-action policy. arXiv preprint arXiv:2503.19757. — [arXiv:2503.19757](https://arxiv.org/abs/2503.19757) [find](https://scholar.google.com/scholar?q=Zhi+Hou%2C+Tianyi+Zhang%2C+Yuwen+Xiong%2C+Haonan+Duan%2C+Hengjun+Pu%2C+Ronglei+Tong%2C+Chengyang+Zhao%2C+Xizhou+Zhu%2C+Yu+Qiao%2C+Jifeng+Dai%2C+and+1+others%2E+2025%2E+Dita%3A+Scaling+diffusion+transformer+for+generalist+visio)
14. Oier Mees, Dibya Ghosh, Karl Pertsch, Kevin Black, Homer Rich Walke, Sudeep Dasari, Joey Hejna, Tobias Kreiman, Charles Xu, Jianlan Luo, and 1 others. 2024. Octo: An open-source generalist robot policy. In First Workshop on Vision-Language Models for Navigation and Manipulation at ICRA 2024. [find](https://scholar.google.com/scholar?q=Oier+Mees%2C+Dibya+Ghosh%2C+Karl+Pertsch%2C+Kevin+Black%2C+Homer+Rich+Walke%2C+Sudeep+Dasari%2C+Joey+Hejna%2C+Tobias+Kreiman%2C+Charles+Xu%2C+Jianlan+Luo%2C+and+1+others%2E+2024%2E+Octo%3A+An+open%2Dsource+generalist+robot+polic)
15. Chia-Yu Hung, Qi Sun, Pengfei Hong, Amir Zadeh, Chuan Li, U Tan, Navonil Majumder, Soujanya Poria, and 1 others. 2025. Nora: A small open-sourced generalist vision language action model for embodied tasks. arXiv preprint arXiv:2504.19854. — [arXiv:2504.19854](https://arxiv.org/abs/2504.19854) [find](https://scholar.google.com/scholar?q=Chia%2DYu+Hung%2C+Qi+Sun%2C+Pengfei+Hong%2C+Amir+Zadeh%2C+Chuan+Li%2C+U+Tan%2C+Navonil+Majumder%2C+Soujanya+Poria%2C+and+1+others%2E+2025%2E+Nora%3A+A+small+open%2Dsourced+generalist+vision+language+action+model+for+embodied+t)
16. OpenAI. 2025. Chatgpt. com/. Version 5.2. [find](https://scholar.google.com/scholar?q=OpenAI%2E+2025%2E+Chatgpt%2E+com%2F%2E+Version+5%2E2%2E)
17. Physical Intelligence, Kevin Black, Noah Brown, James Darpinian, Karan Dhabalia, Danny Driess, Adnan Esmail, Michael Equi, Chelsea Finn, Niccolo Fusai, and 1 others. 2025. Pi0.5: A Vision-Language-Action Model with Open-World Generalization. arXiv preprint arXiv:2504.16054. [https://chat.openai](https://chat.openai). [↗](https://chat.openai)
18. Abby O’Neill, Abdul Rehman, Abhiram Maddukuri, Abhishek Gupta, Abhishek Padalkar, Abraham Lee, Acorn Pooley, Agrim Gupta, Ajay Mandlekar, Ajinkya Jain, and 1 others. 2024. Open x-embodiment: Robotic learning datasets and rt-x models: Open x-embodiment collaboration 0. In 2024 IEEE International Conf… [find](https://scholar.google.com/scholar?q=Abby+O%E2%80%99Neill%2C+Abdul+Rehman%2C+Abhiram+Maddukuri%2C+Abhishek+Gupta%2C+Abhishek+Padalkar%2C+Abraham+Lee%2C+Acorn+Pooley%2C+Agrim+Gupta%2C+Ajay+Mandlekar%2C+Ajinkya+Jain%2C+and+1+others%2E+2024%2E+Open+x%2Dembodiment%3A+Robotic+l)
19. Moo Jin Kim, Chelsea Finn, and Percy Liang. 2025. Fine-tuning vision-language-action models: Optimizing speed and success. arXiv preprint arXiv:2502.19645. [find](https://scholar.google.com/scholar?q=Moo+Jin+Kim%2C+Chelsea+Finn%2C+and+Percy+Liang%2E+2025%2E+Fine%2Dtuning+vision%2Dlanguage%2Daction+models%3A+Optimizing+speed+and+success%2E+arXiv+preprint+arXiv%3A2502%2E19645%2E)
20. Karl Pertsch, Kyle Stachowicz, Brian Ichter, Danny Driess, Suraj Nair, Quan Vuong, Oier Mees, Chelsea Finn, and Sergey Levine. 2025. [FAST: Efficient action tokenization for vision-language-action models](https://md.research.elamur.ai/free/en/papers/pertsch-2025-fast-efficient-action-tokenization-vision/). arXiv preprint arXiv:2501.09747.
21. Moo Jin Kim, Karl Pertsch, Siddharth Karamcheti, Ted Xiao, Ashwin Balakrishna, Suraj Nair, Rafael Rafailov, Ethan Foster, Grace Lam, Pannag Sanketi, and 1 others. 2024. Openvla: An open-source vision-language-action model. arXiv preprint arXiv:2406.09246. [find](https://scholar.google.com/scholar?q=Moo+Jin+Kim%2C+Karl+Pertsch%2C+Siddharth+Karamcheti%2C+Ted+Xiao%2C+Ashwin+Balakrishna%2C+Suraj+Nair%2C+Rafael+Rafailov%2C+Ethan+Foster%2C+Grace+Lam%2C+Pannag+Sanketi%2C+and+1+others%2E+2024%2E+Openvla%3A+An+open%2Dsource+vision%2D)
22. Delin Qu, Haoming Song, Qizhi Chen, Yuanqi Yao, Xinyi Ye, Yan Ding, Zhigang Wang, JiaYuan Gu, Bin Zhao, Dong Wang, and 1 others. 2025. Spatialvla: Exploring spatial representations for visual-language-action model. arXiv preprint arXiv:2501.15830. — [arXiv:2501.15830](https://arxiv.org/abs/2501.15830) [find](https://scholar.google.com/scholar?q=Delin+Qu%2C+Haoming+Song%2C+Qizhi+Chen%2C+Yuanqi+Yao%2C+Xinyi+Ye%2C+Yan+Ding%2C+Zhigang+Wang%2C+JiaYuan+Gu%2C+Bin+Zhao%2C+Dong+Wang%2C+and+1+others%2E+2025%2E+Spatialvla%3A+Exploring+spatial+representations+for+visual%2Dlanguage)
23. Bo Li, Yuanhan Zhang, Dong Guo, Renrui Zhang, Feng Li, Hao Zhang, Kaichen Zhang, Peiyuan Zhang, Yanwei Li, Ziwei Liu, and Chunyuan Li. 2024a. [Llava-onevision: Easy visual task transfer](https://md.research.elamur.ai/free/en/papers/li-2024-llava-onevision-easy-visual-task/). Transactions on Machine Learning Research.
24. Qixiu Li, Yaobo Liang, Zeyu Wang, Lin Luo, Xi Chen, Mozheng Liao, Fangyun Wei, Yu Deng, Sicheng Xu, Yizhong Zhang, and 1 others. 2024b. Cogact: A foundational vision-language-action model for synergizing cognition and action in robotic manipulation. arXiv preprint arXiv:2411.19650. — [arXiv:2411.19650](https://arxiv.org/abs/2411.19650) [find](https://scholar.google.com/scholar?q=Qixiu+Li%2C+Yaobo+Liang%2C+Zeyu+Wang%2C+Lin+Luo%2C+Xi+Chen%2C+Mozheng+Liao%2C+Fangyun+Wei%2C+Yu+Deng%2C+Sicheng+Xu%2C+Yizhong+Zhang%2C+and+1+others%2E+2024b%2E+Cogact%3A+A+foundational+vision%2Dlanguage%2Daction+model+for+synergiz)
25. Weiqi Li, Quande Zhang, Ruifeng Zhai, Liang Lin, and Guangrun Wang. 2025. Vla models are more generalizable than you think: Revisiting physical and spatial modeling. arXiv preprint arXiv:2512.02902. — [arXiv:2512.02902](https://arxiv.org/abs/2512.02902) [find](https://scholar.google.com/scholar?q=Weiqi+Li%2C+Quande+Zhang%2C+Ruifeng+Zhai%2C+Liang+Lin%2C+and+Guangrun+Wang%2E+2025%2E+Vla+models+are+more+generalizable+than+you+think%3A+Revisiting+physical+and+spatial+modeling%2E+arXiv+preprint+arXiv%3A2512%2E02902%2E+%E2%80%94)
26. Fanqi Lin, Ruiqian Nai, Yingdong Hu, Jiacheng You, Junming Zhao, and Yang Gao. 2025. Onetwovla: A unified vision-language-action model with adaptive reasoning. arXiv preprint arXiv:2505.11917. — [arXiv:2505.11917](https://arxiv.org/abs/2505.11917) [find](https://scholar.google.com/scholar?q=Fanqi+Lin%2C+Ruiqian+Nai%2C+Yingdong+Hu%2C+Jiacheng+You%2C+Junming+Zhao%2C+and+Yang+Gao%2E+2025%2E+Onetwovla%3A+A+unified+vision%2Dlanguage%2Daction+model+with+adaptive+reasoning%2E+arXiv+preprint+arXiv%3A2505%2E11917%2E+%E2%80%94+%5BarXi)
27. Bo Liu, Yifeng Zhu, Chongkai Gao, Yihao Feng, Qiang Liu, Yuke Zhu, and Peter Stone. 2023. Libero: Benchmarking knowledge transfer for lifelong robot learning. Advances in Neural Information Processing Systems, 36:44776–44791. [find](https://scholar.google.com/scholar?q=Bo+Liu%2C+Yifeng+Zhu%2C+Chongkai+Gao%2C+Yihao+Feng%2C+Qiang+Liu%2C+Yuke+Zhu%2C+and+Peter+Stone%2E+2023%2E+Libero%3A+Benchmarking+knowledge+transfer+for+lifelong+robot+learning%2E+Advances+in+Neural+Information+Processing)
28. Songming Liu, Lingxuan Wu, Bangguo Li, Hengkai Tan, Huayu Chen, Zhengyi Wang, Ke Xu, Hang Su, and Jun Zhu. 2024. Rdt-1b: a diffusion foundation model for bimanual manipulation. arXiv preprint arXiv:2410.07864. — [arXiv:2410.07864](https://arxiv.org/abs/2410.07864) [find](https://scholar.google.com/scholar?q=Songming+Liu%2C+Lingxuan+Wu%2C+Bangguo+Li%2C+Hengkai+Tan%2C+Huayu+Chen%2C+Zhengyi+Wang%2C+Ke+Xu%2C+Hang+Su%2C+and+Jun+Zhu%2E+2024%2E+Rdt%2D1b%3A+a+diffusion+foundation+model+for+bimanual+manipulation%2E+arXiv+preprint+arXiv%3A24)
29. Qwen Team. 2026. Qwen3.5: Towards native multimodal agents. [find](https://scholar.google.com/scholar?q=Qwen+Team%2E+2026%2E+Qwen3%2E5%3A+Towards+native+multimodal+agents%2E)
30. Moritz Reuss, Ömer Erdinç Yağmurlu, Fabian Wenzel, and Rudolf Lioutikov. 2024. Multimodal diffusion transformer: Learning versatile behavior from multimodal goals. arXiv preprint arXiv:2407.05996. — [arXiv:2407.05996](https://arxiv.org/abs/2407.05996) [find](https://scholar.google.com/scholar?q=Moritz+Reuss%2C+%C3%96mer+Erdin%C3%A7+Yag%CC%86murlu%2C+Fabian+Wenzel%2C+and+Rudolf+Lioutikov%2E+2024%2E+Multimodal+diffusion+transformer%3A+Learning+versatile+behavior+from+multimodal+goals%2E+arXiv+preprint+arXiv%3A2407%2E05996%2E+%E2%80%94+)
31. Shuhan Tan, Kairan Dou, Yue Zhao, and Philipp Krähenbühl. 2025. Interactive post-training for vision-language-action models. arXiv preprint arXiv:2505.17016. — [arXiv:2505.17016](https://arxiv.org/abs/2505.17016) [find](https://scholar.google.com/scholar?q=Shuhan+Tan%2C+Kairan+Dou%2C+Yue+Zhao%2C+and+Philipp+Kr%C3%A4henb%C3%BChl%2E+2025%2E+Interactive+post%2Dtraining+for+vision%2Dlanguage%2Daction+models%2E+arXiv+preprint+arXiv%3A2505%2E17016%2E+%E2%80%94+%5BarXiv%3A2505%2E17016%5D%28https%3A%2F%2Farxiv%2Eorg%2Fabs)
32. Gemma Team, Thomas Mesnard, Cassidy Hardin, Robert Dadashi, Surya Bhupatiraju, Shreya Pathak, Laurent Sifre, Morgane Rivière, Mihir Sanjay Kale, Juliette Love, and 1 others. 2024. Gemma: Open models based on gemini research and technology. arXiv preprint arXiv:2403.08295. [find](https://scholar.google.com/scholar?q=Gemma+Team%2C+Thomas+Mesnard%2C+Cassidy+Hardin%2C+Robert+Dadashi%2C+Surya+Bhupatiraju%2C+Shreya+Pathak%2C+Laurent+Sifre%2C+Morgane+Rivi%C3%A8re%2C+Mihir+Sanjay+Kale%2C+Juliette+Love%2C+and+1+others%2E+2024%2E+Gemma%3A+Open+models+b)
33. Hongtao Wu, Ya Jing, Chilam Cheang, Guangzeng Chen, Jiafeng Xu, Xinghang Li, Minghuan Liu, Hang Li, and Tao Kong. 2023. Unleashing large-scale video generative pre-training for visual robot manipulation. arXiv preprint arXiv:2312.13139. — [arXiv:2312.13139](https://arxiv.org/abs/2312.13139) [find](https://scholar.google.com/scholar?q=Hongtao+Wu%2C+Ya+Jing%2C+Chilam+Cheang%2C+Guangzeng+Chen%2C+Jiafeng+Xu%2C+Xinghang+Li%2C+Minghuan+Liu%2C+Hang+Li%2C+and+Tao+Kong%2E+2023%2E+Unleashing+large%2Dscale+video+generative+pre%2Dtraining+for+visual+robot+manipulati)
34. Zhihao Zhan, Jiaying Zhou, Likui Zhang, Qinhan Lv, Hao Liu, Jusheng Zhang, Weizheng Li, Ziliang Chen, Tianshui Chen, Ruifeng Zhai, Keze Wang, Liang Lin, and Guangrun Wang. 2025. E0: Enhancing generalization and fine-grained control in vla models via tweedie discrete diffusion. arXiv preprint arXiv:2… — [arXiv:2511.21542](https://arxiv.org/abs/2511.21542) [find](https://scholar.google.com/scholar?q=Zhihao+Zhan%2C+Jiaying+Zhou%2C+Likui+Zhang%2C+Qinhan+Lv%2C+Hao+Liu%2C+Jusheng+Zhang%2C+Weizheng+Li%2C+Ziliang+Chen%2C+Tianshui+Chen%2C+Ruifeng+Zhai%2C+Keze+Wang%2C+Liang+Lin%2C+and+Guangrun+Wang%2E+2025%2E+E0%3A+Enhancing+generali)
35. W v , W l ∈ R d are the projection weights for the visual and linguistic modalities, respectively. Qingqing Zhao, Yao Lu, Moo Jin Kim, Zipeng Fu, Zhuoyang Zhang, Yecheng Wu, Zhaoshuo Li, Qianli Ma, Song Han, Chelsea Finn, and 1 others. 2025. Cot-vla: Visual chain-of-thought reasoning for vision-lang… [find](https://scholar.google.com/scholar?q=W+v+%2C+W+l+%E2%88%88+R+d+are+the+projection+weights+for+the+visual+and+linguistic+modalities%2C+respectively%2E+Qingqing+Zhao%2C+Yao+Lu%2C+Moo+Jin+Kim%2C+Zipeng+Fu%2C+Zhuoyang+Zhang%2C+Yecheng+Wu%2C+Zhaoshuo+Li%2C+Qianli+Ma%2C+So)
36. Ruijie Zheng, Yongyuan Liang, Shuaiyi Huang, Jianfeng Gao, Hal Daumé III, Andrey Kolobov, Furong Huang, and Jianwei Yang. 2024. Tracevla: Visual trace prompting enhances spatial-temporal awareness for generalist robotic policies. arXiv preprint arXiv:2412.10345. — [arXiv:2412.10345](https://arxiv.org/abs/2412.10345) [find](https://scholar.google.com/scholar?q=Ruijie+Zheng%2C+Yongyuan+Liang%2C+Shuaiyi+Huang%2C+Jianfeng+Gao%2C+Hal+Daum%C3%A9+III%2C+Andrey+Kolobov%2C+Furong+Huang%2C+and+Jianwei+Yang%2E+2024%2E+Tracevla%3A+Visual+trace+prompting+enhances+spatial%2Dtemporal+awareness+for)
37. Jiaying Zhou, Zhihao Zhan, Ruifeng Zhai, Qinhan Lyu, Hao Liu, Keze Wang, Liang Lin, and Guangrun Wang. 2026. Tag: Target-agnostic guidance for stable object-centric inference in vision-language-action models. arXiv preprint arXiv:2603.24584. — [arXiv:2603.24584](https://arxiv.org/abs/2603.24584) [find](https://scholar.google.com/scholar?q=Jiaying+Zhou%2C+Zhihao+Zhan%2C+Ruifeng+Zhai%2C+Qinhan+Lyu%2C+Hao+Liu%2C+Keze+Wang%2C+Liang+Lin%2C+and+Guangrun+Wang%2E+2026%2E+Tag%3A+Target%2Dagnostic+guidance+for+stable+object%2Dcentric+inference+in+vision%2Dlanguage%2Daction)
38. Xueyang Zhou, Yangming Xu, Guiyao Tie, Yongchao Chen, Guowen Zhang, Duanfeng Chu, Pan Zhou, and Lichao Sun. 2025. Libero-pro: Towards robust and fair evaluation of vision-language-action models beyond memorization. arXiv preprint arXiv:2510.03827. — [arXiv:2510.03827](https://arxiv.org/abs/2510.03827) [find](https://scholar.google.com/scholar?q=Xueyang+Zhou%2C+Yangming+Xu%2C+Guiyao+Tie%2C+Yongchao+Chen%2C+Guowen+Zhang%2C+Duanfeng+Chu%2C+Pan+Zhou%2C+and+Lichao+Sun%2E+2025%2E+Libero%2Dpro%3A+Towards+robust+and+fair+evaluation+of+vision%2Dlanguage%2Daction+models+beyond)
39. Brianna Zitkovich, Tianhe Yu, Sichun Xu, Peng Xu, Ted Xiao, Fei Xia, Jialin Wu, Paul Wohlhart, Stefan Welker, Ayzaan Wahid, and 1 others. 2023. Rt-2: Vision-language-action models transfer web knowledge to robotic control. In Conference on Robot Learning, pages 2165–2183. PMLR. [find](https://scholar.google.com/scholar?q=Brianna+Zitkovich%2C+Tianhe+Yu%2C+Sichun+Xu%2C+Peng+Xu%2C+Ted+Xiao%2C+Fei+Xia%2C+Jialin+Wu%2C+Paul+Wohlhart%2C+Stefan+Welker%2C+Ayzaan+Wahid%2C+and+1+others%2E+2023%2E+Rt%2D2%3A+Vision%2Dlanguage%2Daction+models+transfer+web+knowled)

*Updated 2 November 2026*