DART-GUI: Fully Decoupled Reinforcement Learning with Adaptive Data Curation for GUI Agents. Enable JavaScript to explore interactive demos, or read the paper.