Rerun's New Plugin Lets Robots Record and Train AI in One Step
Could cut hours of robot-data wrangling down to minutes — here's how.
Robots are constantly recording data — cameras, sensors, movement, everything. Rerun, a company that builds tools for viewing that data, just released an experimental plugin that lets a popular robot recording tool (rosbag2) save files straight into Rerun's own format, called RRD. With one command, a robot's recording becomes both a standard rosbag and a Rerun recording at the same time. You can watch it live while recording, then search it afterward.
Why does that matter? Today, engineers usually record robot data, then copy or convert it into another tool before they can analyze it or train AI on it. That's slow, and robot data gets huge fast. RRD stores data like a well-organized spreadsheet, so a search only reads the columns and time ranges it actually needs. The same setup plugs directly into PyTorch, a popular AI training tool, so engineers can train a model straight from their recordings — no export step, no waiting.
The catch: this is experimental. Rerun says it's early, issues and bug reports are welcome, and the details are still being worked out. It will be covered in a talk at ROSCon, a robotics conference, in a few weeks. This is also a niche tool — if you don't work with robots or sensor data, it won't change your day.
The bigger picture is that robot data pipelines are the unglamorous plumbing behind self-driving cars, warehouse robots, and delivery drones. Every hour engineers save wrangling files is an hour spent making robots work better and safer. Faster tools mean the robots already heading into your neighborhood get smarter sooner.
- Two companies' formats become one file: a robot recording now works as both a standard rosbag and a Rerun recording
- Engineers can train AI models directly from recordings, skipping the usual slow copy-and-convert step
- Stores data like a spreadsheet, so searches read only the columns and time ranges needed
- It's experimental and currently aimed at robotics developers, not everyday users
Why It Matters
Faster robot data pipelines mean smarter, safer robots sooner — and less grunt work for the engineers who build them.