Robocasa Pretrain Human300 V4: Human-Robot Interaction Demonstrations

300 human demonstrations for robot pretraining, created using the LeRobot framework. The dataset includes video observations from a robot agent's perspective, with frames captured at 20 frames per second and a resolution of 256x256 pixels. It was uploaded by author 'pepijn223' and last updated on June 5, 2026.

Use Cases

• Training robot policies via imitation learning based on human demonstration videos.
• Pretraining visual representations for robotic manipulation tasks using the provided video observations.
• Benchmarking robot learning algorithms on a standardized set of human teleoperation data.
• Developing multi-task robotic agents using the diverse set of demonstrations likely contained in the dataset.

Strengths

• Video data is standardized at 256x256 resolution and 20 FPS, ensuring consistency.
• Dataset is designed for pretraining within a specific robotics framework (LeRobot), suggesting integration support.

Limitations

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

Provenance

Source

LeRobot framework

Collection Method

Human demonstrations

Freshness

Last updated 2026-06-05 12:13:25; freshness should be verified.