Robotics · raw logs → training dataset

robotdata-pipeline

GitHub Code ↗ ← Back to hub

A raw-log-to-dataset pipeline with a contract that returns a failing exit code when the output should not ship.

Public · synthetic demo

The problem

A pipeline that always writes output can quietly manufacture bad training data.

The contract must be allowed to stop production, not merely add a warning beside a broken dataset.

The money shot

The broken run stops CI

Contract verdicts

run	contract	exit code
clean	PASS	0
broken	FAIL	1

Exit code 1 is the product behavior: CI can stop the build before a bad dataset is released.

Clean output

4dataset episodes

44dataset frames

LeRobotrecommended format

Honest limitations

This is data infrastructure around robot programs. It does not build, control, train, or evaluate a robot; it transforms supplied logs and enforces a declared dataset contract.

This page reports the supplied sample output and its stated limits. Inspect the code and evidence ↗