Add robotics task category and paper link

README.md

@@ -1,6 +1,11 @@
+---
+task_categories:
+- robotics
+---
+
 # WatchAct: A Benchmark for Behavior-Grounded Robot Manipulation
 
-🌐 [Project Page](https://baiqi-li.github.io/watchact_page/)  |  🤗 [Dataset](https://huggingface.co/datasets/BaiqiL/WatchAct)  |  💻 [Code](https://github.com/Baiqi-Li/WatchAct)
+🌐 [Project Page](https://baiqi-li.github.io/watchact_page/)  |  🤗 [Dataset](https://huggingface.co/datasets/BaiqiL/WatchAct)  |  💻 [Code](https://github.com/Baiqi-Li/WatchAct)  |  📄 [Paper](https://huggingface.co/papers/2606.26443)
 
 WatchAct is a behavior-grounded benchmark for robotic manipulation, where the robot reasons over observed human behavior and a language instruction to perform the corresponding task.
 
@@ -36,4 +41,4 @@ tasks, grouped by capability:
 | Episodic Reasoning | `Restore_Previous_State`, `Task_Continuation`, `Error_Correction`, `Conditional_Execution` |
 
 See the [code repository](https://github.com/Baiqi-Li/WatchAct) for the full
-evaluation pipeline.
+evaluation pipeline.