Robotics dataset validation
I validate robotics datasets so labs stop training on garbage.
Before you spend GPU hours on a policy, find out what's actually in your teleop and demonstration data.
Get in touchWhat we do
SRA Robotics offers independent quality checks on robot learning datasets. Here's what a validation pass covers:
Label & annotation checks
Look for mislabeled, missing, or inconsistent task labels and episode metadata.
Sync & timestamp audits
Check alignment across cameras, joint states, and actions, and flag drift, gaps, and dropped frames.
Corrupt & duplicate episodes
Find truncated, unreadable, or near-duplicate episodes that quietly skew training.
Sensor calibration gaps
Spot missing or inconsistent intrinsics/extrinsics and calibration changes across sessions.
Who it's for
- Robotics labs training manipulation or humanoid policies on teleop and demonstration data
- Robotics startups collecting their own datasets and wanting a second set of eyes before training
- Teams merging datasets from multiple robots, rigs, or collection sessions
Founder
Sut Ring AungFounder, SRA Robotics
Contact
Have a dataset you're not sure about? Send a short note about what you're collecting and how.
[email protected]