VLABench Primitive Tasks: Robot Learning Videos in LeRobot Format

A dataset organized in the LeRobot v3.0 format for integrating VLABench into the LeRobot framework. It stores visual observations in a video-compressed format, which likely provides advantages in storage efficiency and data loading performance compared to previous image-based versions. The dataset was authored by VLABench and last updated on 2026-04-21.

Use Cases

• Training robot manipulation policies based on video observations.
• Benchmarking robot learning algorithms on primitive tasks.
• Studying the efficiency of video-compressed vs. image-based observation formats for robot learning.

Strengths

• Uses a video-compressed format for visual observations, which the description states provides significant advantages in storage efficiency and data loading performance.
• Officially integrated into the LeRobot framework, suggesting a standardized format for compatibility.

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.

Provenance

Source

VLABench

Freshness

Last updated 2026-04-21 01:39:57; freshness should be verified.