AgentR1/Claw-R1

Registry 信息

项目简介

Claw-R1 是一个开源项目，通过先进的代理强化学习（Agentic RL）能力增强 OpenClaw，实现基于代理的系统中的智能自适应行为。

要点

• 将代理强化学习集成到 OpenClaw 中，实现自主决策
• 基于 Python 构建，易于使用和扩展
• 专为自适应和基于学习的代理行为设计
• 开源且由社区驱动开发

使用场景

• 在模拟环境中训练智能代理
• 通过基于强化学习的控制策略增强 OpenClaw
• 代理强化学习研究
• 构建自适应自动化系统

README 摘要

<h1 align="center"> Claw-R1: The Data Foundation for <br> Agentic Reinforcement Learning </h1> <p align="center"> <a href="https://agentr1.github.io/"><img src="https://img.shields.io/badge/Project-Home-orange.svg" alt="Project Home"></a> <a href="https://github.com/AgentR1/Claw-R1/stargazers"><img src="https://img.shields.io/github/stars/AgentR1/Claw-R1" alt="GitHub Repo stars"></a> <a href="https://github.com/AgentR1/Claw-R1/network/members"><img src="https://img.shields.io/github/forks/AgentR1/Claw-R1" alt="GitHub forks"></a> <a href="https://agentr1.github.io/Claw-R1/"><img src="https://img.shields.io/badge/docs-latest-blue.svg" alt="Docs"></a> </p> <p align="center"><img src="./assets/logo.jpeg" width="600px" alt="Claw-R1 Logo" /></p> ## News - **[2026.06]** **Claw-R1 Demo Technical Report Updated.** We updated the Claw-R1 Demo technical report: [https://arxiv.org/abs/2606.09138](https://arxiv.org/abs/2606.09138). - **[2026.06]** **Claw-R1 Dashboard Released.** The live dashboard adds Agentic RL data lifestyle management: collection monitoring, step-level representation, curation signals, prefix-tree optimization preview, and training consumption tracking. - **[2026.04]** 🌲 **Prefix Tree Merge for Agentic RL Training.** A new algorithm that deduplicates shared prefix computation in multi-step agent training via prefix tree packing + FlexAttention. Currently under testing on the [`prefix-tree-merge`](https://github.com/AgentR1/Claw-R1/tree/prefix-tree-merge) branch. See [documentation](https://agentr1.github.io/Claw-R1/components/prefix-tree-merge/). - **[2026.04]** 📚 **RL Training Internals Tutorial.** A comprehensive tutorial covering core RL concepts (Reward / Value / Advantage / Return / Loss), PPO & GRPO algorithms, and Claw-R1's step-level agent

话题