ROBOTICS DATA, EVALUATED
PUBLIC REPORT · SNAPSHOT 28 SEP 2026
76
/100
lerobot/aloha_static_towel
Technical readiness
For the evaluated episodes and scored checks.
Episodes evaluated
50
Pass critical checks
39 / 50
Blocking episodes
11 / 50
Passing-episode quality
96.9 / 100
Check types
Engine 0.6.4 · Dataset revision 13ad96f
FINDINGS BY DEVICE
Robot arm · Stuck signal · 12 episodes
Gripper · Missing samples · 5 episodes
Wrist IMU · Timing gaps · 3 episodes
SUPPORTED FORMATS & INPUTS
Work with local datasets or supported datasets on Hugging Face.
Some formats require optional packages. Evaluation coverage varies by input.
BUILT FOR PROGRESS
Use evidence from evaluation to guide review, collection, and the next step your team takes.
01
Accelerate dataset review
Direct your team to the episodes and signals that need attention, with evidence behind each finding.
02
Make every collection count
Use early quality checks to guide investigation and your next recording session.
03
Build a repeatable process
Bring consistent evaluation to each new batch, with documented findings your team can review together.
SEE THE EVIDENCE
Inspect affected episodes, review findings, and understand evaluation coverage before deciding what to investigate next.
Episodes needing attention
See how many evaluated episodes have blocking findings. Review the evidence before deciding whether to repair, recollect or exclude data.
Measurement scope
Distinguish checks used in the score from measurements reported for review. Unavailable checks and missing coverage are visible limits, not passing results.
Report provenance
Check the dataset revision and engine version before comparing results. Changes to the evaluator can change the score even when the data stays the same.
Technical readiness describes the evaluated episodes and checks. Image/video quality and task success are not evaluated. These counts do not establish full-dataset coverage, and a high score does not guarantee model performance.
04 / CONTACT
Collecting robot demonstrations regularly? Tell us about your formats, recording setup and review workflow. We’re working with teams to make recurring data checks more useful.
KALANOS / DATA QUALITY FOR PHYSICAL AI
© 2026