Uni-GUI-Desktop-1: Large-Scale Desktop GUI Agent Trajectories

Uni-GUI-Desktop-1 is a large-scale dataset of 2,685 desktop GUI agent trajectories containing approximately 36,000 total steps. It was created by UI-MOPD and serves as training data for multi-platform on-policy distillation in continual GUI agent learning. The dataset covers 10 application categories, with screen coordinates normalized to a [0, 999] range on a 1920x1080 desktop platform.

Use Cases

• Training reinforcement learning agents for desktop automation based on recorded GUI interaction trajectories.
• Benchmarking GUI agent performance across different application categories mentioned in the description.
• Studying human-computer interaction patterns from normalized screen coordinate data.
• Developing continual learning methods for agents that operate across multiple software platforms.

Strengths

• Contains 2,685 distinct agent trajectories, providing a substantial base for training.
• Includes approximately 36,000 interaction steps, offering detailed sequential data.
• Covers 10 different application categories, suggesting diversity in task types.
• Uses a consistent normalized coordinate system [0, 999] for screen interactions.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

UI-MOPD

Collection Method

Likely collected from automated or simulated interactions with desktop applications.

Freshness

Last updated 2026-07-04 11:08:04; freshness should be verified.