Underwater Robots' Data Has Hidden Traps — New Guide Exposes Them

Underwater robots like AUVs (autonomous underwater vehicles) and ROVs (remotely operated vehicles) are used to map the seafloor, inspect pipelines, and explore caves. These robots record every ping from their sonar, every twist of their gyroscopes, and every bit of navigation data. But this data can be messy and even misleading if you don't know what to look for.

One marine robotics hobbyist, Bruce Krogman, dug into how this data is structured in ROS 2 (a popular toolkit for robot software) and shared his findings. He created two free documents — an animated field guide and a glossary of 34 terms — to help people understand what the numbers actually mean. To prove his point, he analyzed a real 32-minute recording of a cave survey. The recording held over 661,000 messages, and his deep dive uncovered several unexpected quirks.

For example, the vehicle's position tracker never once recorded forward speed, even though it always left that field blank. The speed had to be recovered from a different sensor. Another sensor reported depth as "minus two" whenever it lost sight of the seafloor, which would throw off any average. And once in a while, the orientation data would suddenly flip upside-down, making the submarine appear to do a backflip without moving. If you just glance at the charts, these blips look like real motion — but they're not.

Krogman's guide is aimed at people who work with robot data, but it matters to the rest of us too. When autonomous submarines mapped the Titanic, or when robots search for lost divers, the decisions made from this kind of data can be life-or-death. If rescue teams or data scientists misinterpret a zero speed or a fake depth, they could waste expensive hours — or worse. His work is a reminder that raw data is never truly raw; it comes with a story.