Observed arrival · 2026-09-30
Kinesift checks robot-learning datasets before training
Kinesift checks LeRobot-format datasets for recording and metadata problems that can undermine robot-learning runs.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- Robot-learning teams working with demonstration datasets
- Worth noticing
- Supports LeRobot dataset formats v2.0, v2.1, and v3.0; each finding is paired with a standalone script that reads the files directly.
Field notes
The workflow distinguishes public data, submitted by Hugging Face dataset ID, from private data checked on the customer's own machine. Listed checks include timestamp and video gaps, joint-limit violations, near-duplicates, and metadata fields that disagree with the files. The site says findings include affected episodes and a standalone script for inspecting the underlying data. ROS 2 recording checks are explicitly marked as in development, not currently available.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
✦PrettyNotable craft visible
◎NicheUnusually specific use
One card from the complete issue