# robot-data-diff: CLI tool tracks changes in LeRobot datasets

> **Key takeaway:** A new open-source CLI tool called robot-data-diff compares revisions of LeRobot datasets to help identify whether policy retraining failures stem from code or data changes.

**Published:** 2026-10-19T16:16:05+00:00 
**Source:** ROS Robot Operating System News 
**Category:** robotics-science 
**Canonical URL:** https://fosswire.org/news/robot-data-diff-cli-tool-tracks-changes-in-lerobot-datasets.html 

## Summary
robot-data-diff is a new Apache-2.0 licensed CLI tool that diffs any two revisions of LeRobot datasets, either from local directories or Hugging Face repo IDs. It supports LeRobot v2.1 and v3.0 formats and generates per-episode statistics including frame counts, action stats, idle ratios, and action spikes, plus an HTML report. The tool runs CPU-only without downloading video files, requiring no prior tracking setup or commit workflow.

## How robot-data-diff works
- Compares two dataset revisions directly via local paths or Hugging Face repo IDs
- Supports LeRobot dataset formats v2.1 and v3.0
- Reports episodes added, removed, or modified with per-episode stat deltas
- Tracks frame counts, action statistics, idle ratios, and action spikes per episode
- Detects task and schema changes between revisions
- Generates an HTML report for visual comparison
- CPU-only operation; no GPU required
- Videos are never downloaded, reducing bandwidth and storage requirements
- No tracking setup or commit workflow needed before comparison

## Why it matters
The tool addresses a common pain point in robot learning workflows where policy retraining degradation is difficult to attribute to code or data changes. By lowering the barrier to dataset versioning analysis, it may encourage better data hygiene practices in the LeRobot and broader ROS-based robotics community.