Projects
Selected open-source frameworks, methods, and resources built and maintained by REAL Lab.
Full list of repositories: github.com/ZJU-REAL.
An open-source social-media agent that runs the whole creator loop: discover trends, plan topics, create posts, audio and video, publish to Xiaohongshu, Douyin, Zhihu, Bilibili and more, then learn from how each piece performed.
HuggingGPTNeurIPS 2023
—An LLM as controller, the Hugging Face hub as executors. The system plans tasks, selects expert models, runs them, and composes the response, turning language into a universal interface for invoking many AI models.
agevalAgent evaluation on one stable core: swap the agent under test and its environment through plugins, run the same dataset anywhere, and share datasets, plugins, and agent packages through the Hub.
Autonomous end-to-end AI research: literature survey, idea generation and review, experiments on real GPU servers, LaTeX writing, and paper review. Every long task runs as a resumable, auditable, human-gated agent voyage.
ClawGUI
In-context agentic reinforcement learning for skill internalization: agents discover, store, and reuse skills as in-context experience, then internalize them into the policy. Substantial gains over standard RL on ALFWorld and Search-QA.
Self-Distilled Agentic Reinforcement Learning with a gating mechanism. The first open-source codebase to unify agentic RL with on-policy self-distillation, and the base for a growing family of follow-up methods.
GUI-G²Gaussian reward modeling for GUI grounding: dense, location-aware rewards replace sparse hit/miss signals and guide policies toward precise, robust grounding.
Semi-online RL simulates online training from offline trajectories, giving MLLM-based GUI agents stronger multi-turn interaction. UI-S1-7B is state of the art among open 7B models on both semi-online and online metrics.
A continuously updated reading list and resource hub for GUI agent research, covering training, grounding, datasets, and benchmarks, with weekly round-ups of newly accepted papers from AAAI, ICLR, NeurIPS, and more.
No projects match this filter.