# StraTA

**Repository Path**: drhippo/StraTA

## Basic Information

- **Project Name**: StraTA
- **Description**: Implementation for the paper "StraTA: Incentivizing Agentic Reinforcement Learning with Strategic Trajectory Abstraction".
- **Primary Language**: Unknown
- **License**: Apache-2.0
- **Default Branch**: main
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2026-05-18
- **Last Updated**: 2026-05-18

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

<h1 align="center">
    StraTA: Incentivizing Agentic Reinforcement Learning with Strategic Trajectory Abstraction
</h1>

<p align="center">
    Implementation for the paper "StraTA: Incentivizing Agentic Reinforcement Learning with Strategic Trajectory Abstraction".
</p>

<p align="center">
    <a href="https://arxiv.org/abs/2605.06642"> <img alt="arXiv Paper" src="https://img.shields.io/badge/arXiv-Paper-red?style=for-the-badge&logo=arxiv&logoColor=red"></a>
    <a href="https://github.com/xxyQwQ/StraTA"> <img alt="GitHub Code" src="https://img.shields.io/badge/GitHub-Code-blue?style=for-the-badge&logo=github&logoColor=blue"></a>
    <a href="https://huggingface.co/collections/xxyQwQ/strata"> <img alt="HuggingFace Model" src="https://img.shields.io/badge/HuggingFace-Model-yellow?style=for-the-badge&logo=huggingface&logoColor=yellow"></a>
</p>

![Teaser](assets/teaser.png)

In this work, we present **Stra**tegic **T**rajectory **A**bstraction (StraTA), a simple framework that introduces an explicit trajectory-level strategy into agentic reinforcement learning (RL). StraTA samples a compact strategy from the initial task state, conditions subsequent actions on that strategy, and trains strategy generation and action execution jointly with a hierarchical GRPO-style rollout design, further enhanced by diverse strategy rollout and critical self-judgment.

## 📰 News

- **[2026/05/04]** The preprint of our paper is available on [arXiv](https://arxiv.org/abs/2605.06642), with code released on [GitHub](https://github.com/xxyQwQ/StraTA).

## ⚙️ Setup

Our implementation is based on the [rLLM](https://github.com/rllm-org/rllm) framework. You can follow the [tutorial](examples/setup/README.md) to setup the framework.

## 🚀 Usage

Our implementation supports three environments based on the [AgentGym](https://github.com/WooooDyy/AgentGym) codebase. You should first setup the environment and prepare the dataset. Refer to following tutorials for usage:

- [ALFWorld](examples/alfworld/README.md): a text-based embodied household environment.
- [WebShop](examples/webshop/README.md): a web-based online shopping environment.
- [SciWorld](examples/sciworld/README.md): a text-based scientific experimentation environment.

## 🤗 Models

The model checkpoints are available on [HuggingFace](https://huggingface.co/collections/xxyQwQ/strata). You can download them for evaluation.

## 📜 Citation

Please consider citing our paper if you find it helpful:

```bibtex
@article{xue2026strata,
  title={StraTA: Incentivizing Agentic Reinforcement Learning with Strategic Trajectory Abstraction},
  author={Xue, Xiangyuan and Zhou, Yifan and Wang, Zidong and Tang, Shengji and Torr, Philip and Ouyang, Wanli and Bai, Lei and Yin, Zhenfei},
  journal={arXiv preprint arXiv:2605.06642},
  year={2026}
}
```