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.

What we do

SRA Robotics offers independent quality checks on robot learning datasets. Here's what a validation pass covers:

Look for mislabeled, missing, or inconsistent task labels and episode metadata.

Check alignment across cameras, joint states, and actions, and flag drift, gaps, and dropped frames.

Find truncated, unreadable, or near-duplicate episodes that quietly skew training.

Spot missing or inconsistent intrinsics/extrinsics and calibration changes across sessions.

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

Sut Ring AungFounder, SRA Robotics

Have a dataset you're not sure about? Send a short note about what you're collecting and how.