Spa-Bench/spa-bench-training-teleoperation-1200

Spa-Bench teleoperated training dataset — 1,200 episodes

Open this dataset in the LeRobot visualizer

This is the full-length, five-camera teleoperated demonstration dataset used for Spa-Bench adaptation experiments. The native Hugging Face Data Studio viewer is enabled through the Parquet files declared above.

Dataset summary

Field	Value
Episodes	1,200
Frames	612,733
Duration at 30 FPS	approximately 5.7 hours
Unique instruction strings	321
Task families	6; 200 demonstrations per family
Objects	25 objects across 11 morphological families
Camera streams	middle, wrist, above, left, right
Home-start / recovery demonstrations	883 / 317

The task families are Counting, Ordinal Position, Physical State, Referential Description, Relational Placement, and Relative Size. The benchmark withholds selected concept–argument combinations while retaining exposure to their constituent concepts, objects, and broad manipulation behavior.

Training-design records and the 1,200-row prompt manifest are in spa-bench-data-information.

Dataset structure

This is a LeRobot v3 dataset at 30 FPS. Each frame contains an episode index, frame index, timestamp, task index, six-dimensional observation state, six-dimensional action, and synchronized 480×640 RGB observations. The evaluated policies use the middle and wrist views; all five views are retained here.

Uses, limitations, and safety

This dataset is not representative of general homes or unconstrained manipulation. It uses one SO-101 embodiment, one workspace, one collection setup, a limited object set, and English instructions. Robot data and learned policies can enable unsafe motion; use hardware safeguards and direct human supervision.

Double-blind release note

Author, institution, source-repository, and archival citation details are intentionally omitted during review. They will be restored in the archival release.