Skip to main content
  1. Home
  2. Emerging Tech
  3. News

AI can turn a cloud of scuba bubbles into an early warning for divers

An underwater robot can estimate breathing rates from video, although the research stops short of proving that it can diagnose stress or distress

Add as a preferred source on Google
Nature, Outdoors, Water, Scuba
NEOM / Unsplash

University of Minnesota researchers have taught an AI underwater robot to estimate a diver’s breathing rate by watching the bubbles released with each exhalation.

The contactless system, detailed in a peer-reviewed study published in The International Journal of Robotics Research, works with ordinary camera footage. Divers don’t need to squeeze another sensor into an equipment loadout that is hardly lacking for accessories.

Recommended Videos

Changes in breathing could eventually alert robotic dive partners when something looks wrong. At this stage, the system measures respiration and leaves humans to decide what the reading means.

How the robot read bubbles

The robot’s cameras watch for the bursts of bubbles produced whenever a diver exhales. Its software identifies each burst and converts the pattern into an estimated number of breaths per minute.

Training it required more than pointing a camera into the water and trusting the bubbles to cooperate. Researchers manually labeled thousands of underwater images and matched the footage with audio of regulator exhalations, teaching the system what counted as a breath.

The team collected video and audio in Minnesota lakes and the Caribbean, testing the method across different temperatures and visibility conditions.

What the robot show divers

During field trials, the HREyes system sorted readings into three bands. Fewer than 14 breaths per minute counted as below normal, while more than 20 counted as above normal.

The robot could then display that status to the diver. Think of it as an unusual-reading indicator rather than a proven emergency alarm. It can spot when breathing moves outside the expected range, but the bubbles won’t explain why.

What still needs proving

The study compared the robot’s estimates with human analysis. It didn’t establish whether those readings can reliably identify stress or warn that an emergency is approaching.

Researchers next plan to combine respiration measurements with analysis of the diver’s movement. That second physical signal could give the robot more context. Until then, it can provide a breathing rate, not a diagnosis.

Paulo Vargas
Paulo Vargas is an English major turned reporter turned technical writer, with a career that has always circled back to…
Microsoft wants AI to catch hackers before they even attack
Red team, blue team, green team: Microsoft just gave its AI a full security squad.
Microsoft Project Perception

With AI now capable of writing exploits and running attacks at machine speed, the security tools built for a slower, human-paced world are starting to show their age. Microsoft thinks it has the answer, and it's calling it Project Perception.

So what exactly is Project Perception?

Read more
Nike’s new slide is basically a spa day for your feet
Nike and Hyperice built a slide that literally heats up and vibrates your feet back to life.
nike-air-zoom-hyperslide

Nike just teamed up with Hyperice again, and this time they've built something that feels more like a wellness gadget than a shoe. Meet the Nike Air Zoom Hyperslide, a slide that not only cushions your feet but also heats and vibrates them back to life. It's made less for lounging by the pool and more for helping you recover after training, travel, or just a long day spent standing around.

What is the Nike Hyperslide?

Read more
That Claude chat you shared? Google may have seen it too
Public chat links were indexed by Google, potentially exposing sensitive information shared with the AI.
Claude AI on an iPhone.

If you've ever shared a Claude conversation using a public link, it might be worth checking your privacy settings. Over the weekend, users discovered that publicly shared Claude chats and Artifacts were appearing in Google Search and other search engines through queries such as site:claude.ai/share. The indexed pages reportedly included everything from resumes and business documents to API keys, health information, and other sensitive conversations that users likely never expected to become searchable.

How did this happen?

Read more