Scaling Synthetic Task Generation for Agents via Exploration

AutoPlay is a scalable task-generation pipeline that first explores interactive environments to uncover functionalities and then synthesizes diverse, executable, verifiable tasks grounded in those states. It generates 20k Android tasks and 10k Ubuntu tasks, enabling large-scale post-training and additional RL gains for UI agents without human annotation.

• Moving Beyond Sparse Grounding with Complete Screen Parsing Supervision
  February 15, 2026 · arXiv
• GUIGuard: Toward a General Framework for Privacy-Preserving GUI Agents
  January 26, 2026 · arXiv
• Beyond Clicking: A Step Towards Generalist GUI Grounding via Text Dragging
  November 7, 2025 · arXiv
• VideoAgentTrek: Computer Use Pretraining from Unlabeled Videos
  October 22, 2025 · arXiv
• Scaling Computer‑Use Grounding via User Interface Decomposition and Synthesis
  May 19, 2025 · NeurIPS 2025 Datasets and Benchmarks Track (Spotlight)
• TongUI: Internet-Scale Trajectories from Multimodal Web Tutorials for Generalized GUI Agents
  April 17, 2025 · AAAI 2026