Add model card info and metadata

README.md

@@ -1,6 +1,39 @@
 ---
+license: mit
+pipeline_tag: robotics
 base_model:
 - Gen-Verse/MMaDA-8B-Base
 ---
 
-Paper: https://arxiv.org/abs/2603.25406
+# MMaDA-VLA: Large Diffusion Vision-Language-Action Model with Unified Multi-Modal Instruction and Generation
+
+[**Project Page**](https://yliu-cs.github.io/MMaDA-VLA) | [**Paper**](https://arxiv.org/abs/2603.25406) | [**GitHub**](https://github.com/yliu-cs/MMaDA-VLA)
+
+**MMaDA-VLA** is a fully native pre-trained large diffusion Vision-Language-Action (VLA) model that unifies multi-modal understanding and generation in a single framework. It employs a native discrete diffusion formulation that embeds language, images, and continuous robot controls into one discrete token space.
+
+## Model Summary
+MMaDA-VLA trains a single backbone with masked token denoising to jointly generate a future goal observation and an action chunk in parallel. Iterative denoising enables global, order-free refinement, improving long-horizon consistency while grounding actions in predicted future visual outcomes without the need for auxiliary world models.
+
+### Key Features
+- **Unified Multi-Modal Space:** Embeds text, vision, and robot actions into a single discrete token space.
+- **Discrete Diffusion Formulation:** Utilizes masked token denoising for parallel generation of actions and future states.
+- **Improved Consistency:** Global refinement via denoising reduces long-horizon error accumulation common in autoregressive models.
+
+## Performance
+MMaDA-VLA achieves state-of-the-art performance across major simulation benchmarks:
+- **LIBERO:** 98.0% average success rate.
+- **CALVIN:** 4.78 average task length.
+
+## Citation
+If you find MMaDA-VLA useful for your research, please consider citing the following paper:
+
+```bibtex
+@article{liu2026mmadavla,
+  author    = {Yang Liu and Pengxiang Ding and Tengyue Jiang and Xudong Wang and Minghui Lin and Wenxuan Song and Hongyin Zhang and Zifeng Zhuang and Han Zhao and Wei Zhao and Siteng Huang and Jinkui Shi and Donglin Wang},
+  title     = {{MMaDA-VLA}: Large Diffusion Vision-Language-Action Model with Multimodal Instruction and Generation},
+  journal   = {CoRR},
+  volume    = {abs/2603.25406},
+  year      = {2026},
+  url       = {https://arxiv.org/abs/2603.25406}
+}
+```