Synthux Economy R8 W149: Simulated Desktop Computer-Use Trajectories

Grounded visual computer-use trajectories generated by the SynthUX system. Each record captures one worker's device session inside a simulated company, where goals drive interactions with desktop-simulator apps like Terminal, VS Code, and Slack. The dataset was authored by jacob-valdez and last updated on the Hugging Face platform in June 2026.

Use Cases

• Training imitation learning agents for desktop automation based on low-level mouse/keyboard input sequences.
• Developing models for visual task understanding based on recorded screen video and per-node frames.
• Studying hierarchical task decomposition based on the goal-to-node-tree structure described.
• Benchmarking multimodal agents that must operate across applications like Terminal, Notes, and Mail.

Strengths

• Data includes multimodal recordings of screen video and per-node frames for each session.
• Trajectories are generated through interaction with a suite of real desktop-simulator applications.
• The dataset description specifies a structured, goal-driven node tree for task expansion.

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.
• Data is simulated, which may not fully capture the complexity of real-world user behavior.

Provenance

Source

huggingface

Collection Method

Generated by the SynthUX system within a simulated company environment.

Freshness

Last updated 2026-06-02 09:31:33.