metadata
license: mit
task_categories:
- other
tags:
- reinforcement-learning
- tool-use
- llm-agents
Agent-STAR TravelDataset
This repository contains the synthetic queries and datasets presented in the paper Demystifying Reinforcement Learning for Long-Horizon Tool-Using Agents: A Comprehensive Recipe.
The dataset serves as a long-horizon tool-use testbed for Reinforcement Learning (RL) research, specifically utilizing the TravelPlanner environment.
- Paper: Demystifying Reinforcement Learning for Long-Horizon Tool-Using Agents: A Comprehensive Recipe
- GitHub Repository: WxxShirley/Agent-STAR
Dataset Description
The dataset includes over 17K synthetic queries generated through a three-step pipeline: element sampling, feasibility checking, and LLM back-translation. These queries require agents to iteratively call tools to satisfy multifaceted constraints.
File Information
| Data | Description |
|---|---|
TravelPlanner_Val180.jsonl |
Official TravelPlanner validation set of 180 instances |
TravelTotal_17K.jsonl |
All 17K+ synthetic queries after element sampling, feasibility checking, and back-translation |
Travel_Mixed_1K_RL.jsonl |
Default 1K RL training set with mixed difficulty |
Travel_Easy_1K.jsonl |
Difficulty-specific 1K set (Easy) for controlled experiments |
Travel_Medium_1K.jsonl |
Difficulty-specific 1K set (Medium) for controlled experiments |
Travel_Hard_1K.jsonl |
Difficulty-specific 1K set (Hard) for controlled experiments |
Related Resources
- Travel Database: Agent-STAR-TravelDatabase (Required for environment interactions)
- Models: Agent-STAR Collection
Citation
@misc{wu2026agentstar,
title={Demystifying Reinforcement Learning for Long-Horizon Tool-Using Agents: A Comprehensive Recipe},
author={Xixi Wu and Qianguo Sun and Ruiyang Zhang and Chao Song and Junlong Wu and Yiyan Qi and Hong Cheng},
year={2026},
eprint={2603.21972},
archivePrefix={arXiv},
primaryClass={cs.LG},
url={https://arxiv.org/abs/2603.21972},
}