Robotwin2.0 Fastwam: Preprocessed Robot Manipulation Data for Action Models

A preprocessed version of the RoboTwin/RobotWin 2.0 dataset, formatted for LeRobot v2.1, released to support the paper 'Fast-WAM: Do World Action Models Need Test-time Future Imagination?'. The dataset was prepared by author 'yuanty' and last updated on March 23, 2026. It is intended for training, evaluation, and reproducibility of the associated research project.

Use Cases

• Training robot action models based on the preprocessed RoboTwin/RobotWin data.
• Evaluating world models that incorporate test-time future imagination, as referenced in the paper.
• Reproducing experiments from the 'Fast-WAM' research project.
• Benchmarking imitation learning algorithms on a preprocessed robotics dataset.

Strengths

• Dataset is preprocessed specifically for the LeRobot v2.1 format, which may reduce setup time.
• Explicitly prepared to support training, evaluation, and reproducibility for a specific research paper.
• Last update timestamp (2026-03-23 13:15:28) is provided.

Limitations

• Description metadata is limited; actual data quality, structure, and content require manual inspection after download.
• Row count, column names, and file formats are unknown, which may limit suitability assessment.
• This is a paper-specific processed version, not the official upstream release, which may affect generalizability.

Provenance

Source

Preprocessed from the RoboTwin / RobotWin 2.0 dataset, likely originating from robotics simulation or real-world robot teleoperation.

Collection Method

Processed by the paper authors for compatibility with LeRobot v2.1 format.

Freshness

Last updated 2026-03-23 13:15:28; freshness should be verified.