Factory video in, training data out
The most expensive part of training a humanoid is labeling the data. Khenda does it automatically: upload your operator video, get back structured datasets your models can learn from.
The pipeline
How it works
You upload video
Standard footage from the cameras you already have. No depth sensors, no motion capture suits, no special rig.
We find the actions
The pipeline detects and timestamps every discrete action: grasps, placements, tool uses, inspections, across multiple workers and angles.
We extract 3D motion
From plain 2D video we infer joint angles, hand trajectories, and contact points, the spatial data robot policies learn from.
You get a dataset
Structured, labeled, and packaged for Vision-Language-Action training. Ready to drop into your pipeline.
What's in the dataset
See it run on your footage
Send us a sample of factory video and we'll turn it into a structured, training-ready dataset so you can see the output format and quality firsthand.
Book a Demo