A dataset of human-collected web-navigation skills, where each skill is a trajectory for a low-level task like 'find_and_open' or 'fill_form'. Each example pairs an instruction with a sequence of webpage screenshots and the corresponding agent actions such as clicks, typing, and scrolling. The dataset was created by AllenAI and last updated on March 24, 2026.
Use Cases
- Train web automation agents based on the paired instructions and action trajectories.
- Benchmark agent performance on low-level tasks like 'find_and_open' or 'fill_form'.
- Develop models for instruction-following in visual web environments based on the screenshot sequences.
- Study human-computer interaction patterns from the recorded click, typing, and scrolling actions.
Strengths
- Pairs textual instructions with visual sequences of webpage screenshots.
- Includes corresponding agent actions like clicks, typing, and scrolling for each trajectory.
- Focuses on low-level, granular web-navigation tasks.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown, which may limit suitability assessment.
- Data may reflect the specific task and user bias inherent to the human collection process.
Provenance
- Source
- AllenAI
- Collection Method
- Human-collected trajectories for low-level web tasks.
- Time Range
- null
- Freshness
- Last updated 2026-03-24 05:34:54; freshness should be verified.
- Geography
- null