Myungkyu

RMBench-preset-luna

RMBench training episodes (9 tasks, 450 episodes) with dense high-level labels produced by GPT-5.6 Luna annotator using only subtask presets without task-specific guidance, including keyframe storage and retrieval annotations.

Downloads574
Episodes450
Likes2

Why This Matters for Physical AI

This dataset provides hierarchical planning annotations with visual and textual memory operations that enable training of high-level planners for multi-step manipulation tasks using minimal task-specific guidance.

Technical Profile

Modalities
rgbproprioception
Action Space
joint_positions
Environment
lab
Task Types
manipulationpick_and_placerearrangement
Episodes
450
Data Format
LeRobot
Annotation Types
language_instructionsaction_labelskeyframe_labelsmemory_annotations
Part of the RMBench family

Access

Need custom rgb data?

Claru builds purpose-built datasets for lab applications with dense human annotations and quality assurance.

Request a Sample Pack

Related Datasets