Open-Travel

# Open-Travel

[**Project Page**](https://tongyi-agent.github.io/blog/arenarl/) | [**Paper**](https://huggingface.co/papers/2601.06487) | [**Code**](https://github.com/Alibaba-NLP/qqr)

This directory contains the **RL Training Set** and the **Test Set** (categorized by subtask) for the Open-Travel domain.

## Overview

In the Open-Travel domain, the agent is required to help users accomplish itinerary planning subtasks. These tasks emphasize multi-constraint reasoning, multi-tool coordination, and personalized preferences intertwined with user-specific constraints (e.g., budget limits, time windows, traveling parties, and preference profiles).

## Dataset

### Statistics

| Split | Samples | Description |
| :-------------- | :-------- | :--------------------------------------------- |
| **RL Training** | **1,626** | Used for Reinforcement Learning (RL) training. |
| **Test** | **250** | Contains 5 subtask files (50 samples each). |
| **Total** | **1,876** | |

### Files

* [`train.jsonl`](train/train.jsonl)
* Contains **1,626** RL training samples.
* [`test/`](test/)
* Contains **250** samples in total, evenly distributed across five distinct subtasks:

| File Name | Samples | Task Type | Description |
| :-------------------------------------------------------- | :------ | :-------- | :------------------------------------------------ |
| [`search_around.jsonl`](test/search_around.jsonl) | **50** | Search | Nearby point-of-interest (POI) search. |
| [`direction.jsonl`](test/direction.jsonl) | **50** | Direction | Route planning with multiple specified waypoints. |
| [`compare_itinerary.jsonl`](test/compare_itinerary.jsonl) | **50** | Compare | Transportation-mode comparison. |
| [`one_day_travel.jsonl`](test/one_day_travel.jsonl) | **50** | 1-Day | One-day trip planning in a single city. |
| [`multi_day_travel.jsonl`](test/multi_day_travel.jsonl) | **50** | M-Day | Multi-day trip planning (Generalization task). |

## License
The dataset files listed in this directory are licensed under the [Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0)](https://creativecommons.org/licenses/by-nc/4.0/).

## 📚 Citation

```bibtex
@misc{zhang2026arenarlscalingrlopenended,
title={ArenaRL: Scaling RL for Open-Ended Agents via Tournament-based Relative Ranking},
author={Qiang Zhang and Boli Chen and Fanrui Zhang and Ruixue Ding and Shihang Wang and Qiuchen Wang and Yinfeng Huang and Haonan Zhang and Rongxiang Zhu and Pengyong Wang and Ailin Ren and Xin Li and Pengjun Xie and Jiawei Liu and Ning Guo and Jingren Zhou and Zheng-Jun Zha},
year={2026},
eprint={2601.06487},
archivePrefix={arXiv},
primaryClass={cs.LG},
url={https://arxiv.org/abs/2601.06487},
}
```

# Open-Travel
[**项目页面**](https://tongyi-agent.github.io/blog/arenarl/) | [**论文**](https://huggingface.co/papers/2601.06487) | [**代码**](https://github.com/Alibaba-NLP/qqr)

本目录包含面向Open-Travel领域的**强化学习训练集**与**按子任务分类的测试集**。

## 概述
在Open-Travel领域中，AI智能体需协助用户完成行程规划子任务。此类任务核心强调多约束推理、多工具协同，以及将用户个性化偏好与专属约束（如预算限制、时间窗口、出行人员构成与偏好画像）相结合的综合处理能力。

## 数据集
### 统计信息
| 划分方式 | 样本数量 | 描述 |
| :-------------- | :-------- | :--------------------------------------------- |
| **强化学习训练** | **1,626** | 用于强化学习（Reinforcement Learning, RL）训练。 |
| **测试集** | **250** | 包含5个子任务文件，每个文件含50个样本。 |
| **总计** | **1,876** | |

### 文件结构
* [`train.jsonl`](train/train.jsonl)
* 包含**1,626**个强化学习训练样本。
* [`test/`](test/)
* 总计包含250个样本，均匀分布于5个不同的子任务：

| 文件名 | 样本数量 | 任务类型 | 描述 |
| :-------------------------------------------------------- | :------ | :-------- | :------------------------------------------------ |
| [`search_around.jsonl`](test/search_around.jsonl) | **50** | 搜索 | 周边兴趣点（Point-of-Interest, POI）搜索。 |
| [`direction.jsonl`](test/direction.jsonl) | **50** | 路线规划 | 含多个指定途经点的路线规划。 |
| [`compare_itinerary.jsonl`](test/compare_itinerary.jsonl) | **50** | 行程对比 | 交通方式对比。 |
| [`one_day_travel.jsonl`](test/one_day_travel.jsonl) | **50** | 单日行程 | 单个城市内的单日行程规划。 |
| [`multi_day_travel.jsonl`](test/multi_day_travel.jsonl) | **50** | 多日行程 | 多日行程规划（泛化任务）。 |

## 许可证
本目录下的数据集文件采用[知识共享署名-非商业性使用4.0国际许可协议（CC BY-NC 4.0）](https://creativecommons.org/licenses/by-nc/4.0/)进行授权。

## 📚 引用

bibtex
@misc{zhang2026arenarlscalingrlopenended,
title={ArenaRL: Scaling RL for Open-Ended Agents via Tournament-based Relative Ranking},
author={Qiang Zhang and Boli Chen and Fanrui Zhang and Ruixue Ding and Shihang Wang and Qiuchen Wang and Yinfeng Huang and Haonan Zhang and Rongxiang Zhu and Pengyong Wang and Ailin Ren and Xin Li and Pengjun Xie and Jiawei Liu and Ning Guo and Jingren Zhou and Zheng-Jun Zha},
year={2026},
eprint={2601.06487},
archivePrefix={arXiv},
primaryClass={cs.LG},
url={https://arxiv.org/abs/2601.06487},
}