Alibaba-DAMO-Academy/RynnVLA-002

Model

stars

commits

linked in READMEs

Nov 22, 2025

updated

safetensors

README

RynnVLA-002: A Unified Vision-Language-Action and World Model

If our project helps you, please give us a star ⭐ on GitHub to support us. 🙏🙏

🌟 Introduction

RynnVLA-002 is an autoregressive action world model that unifies action and image understanding and generation. RynnVLA-002 intergrates Vision-Language-Action (VLA) model (action model) and world model in one single framework. Compared to WorldVLA, RynnVLA-002 adds a continous Action Transformer, wrist camera input and generation, and state input. RynnVLA-002 achieves 97.4% success rate on LIBERO benchmark.

Model Zoo

VLA Model (256 * 256)

Model	HF Link	Continous Action SR (%)	Discrete Action SR (%)
LIBERO-Spatial	Alibaba-DAMO-Academy/RynnVLA-002/VLA_model_256/libero_spatial	99.0	94.2
LIBERO-Object	Alibaba-DAMO-Academy/RynnVLA-002/VLA_model_256/libero_object	99.8	96.8
LIBERO-Goal	Alibaba-DAMO-Academy/RynnVLA-002/VLA_model_256/libero_goal	96.4	94.6
LIBERO-Long	Alibaba-DAMO-Academy/RynnVLA-002/VLA_model_256/libero_10	94.4	87.6

World Model (512 * 512)

Goal	HF Link	FVD↓	PSNR↑	SSIM↑	LPIPS↓
World Model	Alibaba-DAMO-Academy/RynnVLA-002/World_model_512/libero_goal	370.0	22.25	77.84	19.70
Action World Model	Alibaba-DAMO-Academy/RynnVLA-002/Action_World_model_512/libero_goal	336.8	22.13	78.13	19.43

Object	HF Link	FVD↓	PSNR↑	SSIM↑	LPIPS↓
World Model	Alibaba-DAMO-Academy/RynnVLA-002/World_model_512/libero_object	1141.6	20.31	59.59	27.30
Action World Model	Alibaba-DAMO-Academy/RynnVLA-002/Action_World_model_512/libero_object	877.2	22.18	65.03	22.60

Spatial	HF Link	FVD↓	PSNR↑	SSIM↑	LPIPS↓
World Model	Alibaba-DAMO-Academy/RynnVLA-002/World_model_512/libero_spatial	405.4	22.32	79.15	20.28
Action World Model	Alibaba-DAMO-Academy/RynnVLA-002/Action_World_model_512/libero_spatial	373.1	23.88	82.41	16.33

Long	HF Link	FVD↓	PSNR↑	SSIM↑	LPIPS↓
World Model	Alibaba-DAMO-Academy/RynnVLA-002/World_model_512/libero_10	557.73	18.24	69.16	31.60
Action World Model	Alibaba-DAMO-Academy/RynnVLA-002/Action_World_model_512/libero_10	427.86	19.36	72.19	27.78

License

All assets and code are under the Apache 2.0 license unless specified otherwise.

Citation

If you find the project helpful for your research, please consider citing our paper:

@article{cen2025WorldVLA,
  title={WorldVLA: Towards Autoregressive Action World Model},
  author={Cen, Jun and Yu, Chaohui and Yuan, Hangjie and Jiang, Yuming and Huang, Siteng and Guo, Jiayan and Li, Xin and Song, Yibing and Luo, Hao and Wang, Fan and Zhao, Deli and Chen, Hao},
  journal={arXiv preprint arXiv:},
  year={2025}
}

💡 Other featured projects from our RynnBot family ✨.

RynnVLA-001: A Vision-Language-Action Model Boosted by Generative Priors
Yuming Jiang, Siteng Huang, Shengke Xue, Yaxi Zhao, Jun Cen, Sicong Leng, Jiayan Guo, Kexiang Wang, Kehan Li, Mingxiu Chen, Fan Wang, Deli Zhao, Xin Li

RynnEC: Bringing MLLMs into Embodied World
Ronghao Dang*, Yuqian Yuan*, Yunxuan Mao*, Kehan Li*, Jiangpin Liu, Zhikai Wang, Fan Wang, Deli Zhao, Xin Li