New Free Tool Spots When Robot Training Data Goes Bad

When a robot starts behaving badly after retraining, it's hard to know if the problem is the code or the data. A new free tool called robot-data-diff solves this by comparing two versions of a robot's training dataset. You just point it at two folders or online repositories, and it tells you exactly what changed—like which episodes were added, removed, or modified. This saves countless hours of guesswork.

The tool reports detailed stats for each episode, such as frame counts, action statistics, idle ratios, and action spikes. It also flags changes in tasks or data structure. Then it generates a simple HTML report you can share. It's designed to be lightweight: it runs on a regular CPU (no expensive GPU needed) and never downloads the video files, so it's fast and private.

Installing it is as easy as `pip install robot-data-diff`. It's free and open-source under the Apache-2.0 license, meaning anyone can use or modify it. The creator built it to solve a personal pain point: constantly wondering if a regression was due to code or data. Now, anyone working with robot learning can benefit.

The tool is particularly useful for researchers and hobbyists using the LeRobot framework, which is popular for robot learning. By making data changes transparent, it helps teams debug faster and avoid wasting time on the wrong fixes. While it doesn't catch every possible data issue (the creator is open about its limitations), it's a big step toward more reliable robot training.