Add dataset card and link to paper

README.md

@@ -0,0 +1,42 @@
+---
+task_categories:
+- robotics
+---
+
+# AnyPos: Automated Task-Agnostic Actions for Bimanual Manipulation
+
+This repository contains the dataset for the paper [AnyPos: Automated Task-Agnostic Actions for Bimanual Manipulation](https://huggingface.co/papers/2507.12768).
+
+[**Project Page**](https://embodiedfoundation.github.io/vidar_anypos) | [**Github**](https://github.com/yaofeng1998/Vidar)
+
+## Dataset Description
+
+AnyPos is a unified pipeline that integrates large-scale automated task-agnostic exploration with robust embodiment modeling. By exploring feasible actions of an embodiment, the dataset captures physical consistency and decouples embodiment dynamics from high-level policy learning. This allows for scalability and reuse of action data across different tasks like operating a microwave, folding clothes, and watering plants.
+
+## Dataset Structure
+
+According to the official repository, the data is organized as follows:
+
+- **Videos:** `task_name/episode_idx.mp4` (multi-view video files).
+- **Actions:** `task_name/episode_idx_qpos.pt` (2D tensors containing corresponding actions/joint positions).
+
+For training the associated Video Diffusion Model (Vidar), metadata is formatted as:
+```json
+{
+    "video_path": "{VIDEO_PATH}",
+    "raw_caption": {
+        "long caption": "{PROMPT}"
+    }
+}
+```
+
+## Citation
+
+```bibtex
+@article{anypos2025,
+  title={AnyPos: Automated Task-Agnostic Actions for Bimanual Manipulation},
+  author={...},
+  journal={arXiv preprint arXiv:2507.12768},
+  year={2025}
+}
+```