AndroidControl-CL is a dataset for Android GUI control and continual learning, released with the paper 'CGL: Advancing Continual GUI Learning via Reinforcement Fine-Tuning'. The dataset contains screenshot trajectories, JSONL data files, and Parquet files for training and evaluation, with splits by task category. It was uploaded by the author 'Regimes' to Hugging Face and last updated on April 26,我们发现了一个问题。
Use Cases
- Train reinforcement learning agents for Android GUI control based on screenshot trajectories.
- Benchmark continual learning methods using the provided task category splits.
- Fine-tune vision-language models on structured mobile interface interactions.
- Develop automated testing tools for Android applications.
Strengths
- Includes structured train/test splits by task category, which supports evaluation of generalization.
- Provides data in multiple formats (images, JSONL, Parquet) for different processing needs.
- Dataset is associated with a published research paper, suggesting a defined research purpose.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count and dataset size are unknown, which may limit suitability assessment.
- Description metadata is limited; actual data quality requires manual inspection after download.
Provenance
- Source
- Regimes (author on Hugging Face), associated with the paper 'CGL: Advancing Continual GUI Learning via Reinforcement Fine-Tuning'.
- Collection Method
- Likely gathered from automated interactions with Android applications.
- Time Range
- null
- Freshness
- Last updated 2026-04-26 14:20:03; freshness should be verified.
- Geography
- null