orlando23/failed_agent_trajectory

Mobile Trajectory Verification Data

This dataset is prepared for verifying how LoT agent can help us identify incorrect actions taken in mobile execution trajectory. As mentioned in our paper (see citation 1), We select 52 mobile agent trajectories with specific task goals from instruction- guided executions on MagicWand platform(see citation 2). Each folder has a execution trajectory that includes screenshots along this trajectories, json files describing the details of each execution step.

Citation

If you find it useful for your research and applications, please cite using this BibTeX:

@misc{fan2024readpointedlayoutawaregui,
      title={Read Anywhere Pointed: Layout-aware GUI Screen Reading with Tree-of-Lens Grounding}, 
      author={Yue Fan and Lei Ding and Ching-Chen Kuo and Shan Jiang and Yang Zhao and Xinze Guan and Jie Yang and Yi Zhang and Xin Eric Wang},
      year={2024},
      eprint={2406.19263},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2406.19263}, 
}

@inproceedings{dingenhancing,
  title={Enhancing Mobile" How-to" Queries with Automated Search Results Verification and Reranking},
  author={Ding, Lei and Zhang, Yi and Bheemanpally, Jeshwanth},
  booktitle={The Second Workshop on Generative Information Retrieval}
}