A Robot Sub Finds a Cable Its Map Got Wrong | Subsea Cable Tracking Explained
More than 95% of the world's international data travels through cables on the seafloor, but the maps of where those cables lie can be off by many meters. A robot sub that just follows the map with a camera can fly right past the cable.
Researchers at the University of Southampton, Sonardyne International and the University of Tokyo gave their robot a map that was wrong on purpose, shifted 7.5 meters to the side. It found the cable anyway, followed it, and corrected its map of the whole cable along the way. In the best dive, counting both passes, it photographed 59% of the cable you could see, traveling less than a quarter as far as a full back-and-forth survey.
Cables are hard to find. They're usually buried in water up to about 1.5 km deep, and they're mapped from where the ship laid them, which can be off by 5 to 10% of the water depth.
The idea is to keep a map of the whole cable, including the parts the robot hasn't seen, and correct it from what the camera sees. The robot is Smarty200, the team's 2-meter robot sub, with a camera that points straight down. The map is 100 beads on springs, pulled by sightings. A rule keeps one false sighting from dragging it: three in a row, each within 2 meters of the last. And when the robot loses the cable, the physics of a dragged chain tells it how far the cable could have moved. In the trial, the team set the narrowest search zone at 3 meters to either side, and capped it at about 14 meters to either side. The physics sets how fast it grows in between.
The field trial was in Cawsand Bay, near Plymouth: a 120-meter test cable with a gap standing in for a buried stretch. The team gave the robot a map shifted 7.5 meters to the side and ran three tracking dives. On each one, the robot made a pass along the cable, then turned around and came back, starting from its corrected map.
The team counts only the cable you can see, so they leave out the gap. The results:
- 10 to 30% of the cable on the first passes.
- Up to 55% on the way back, with the corrected map.
- 52 to 59% with both passes; the best dive reached 59%.
- A back-and-forth survey saw all of it, but traveled about 12 times the cable's length. The best dive, counting both passes, traveled less than a quarter as far.
What it missed and why: false sightings, where patches of seabed that looked like cable pulled the robot off the route, and a detector that, checked against the back-and-forth survey's 47,000 photos, was right only about 13% of the time it said "cable". Cable photos were rare in that set, and the team says that partly skews these numbers.
The limits: one short cable, in shallow water, over two days. The map error and the search zone were kept small to suit that short cable. The team says so itself. It's built for exposed cable; a camera can't see a cable under the seabed. The team expects faster processing and newer detectors to improve the results. That part hasn't been tested yet.
For engineers
The cable prior is a factor graph of 100 nodes 1.2 m apart, with prior shape constraints and observation edges, solved with Levenberg–Marquardt in g2o. The AUV's poses aren't optimized, so it's graph SLAM turned around. A consecutive-observation filter (N_obs = 3, d_obs = 2 m) gates every map update. A catenary model of the cable on the seabed sets how fast the search boundary widens away from the last observation, between the trial's settings of σ_min = 3 m and σ_max = 14.14 m; the area is covered with a bounded zigzag. Detection is a semi-supervised GeoCLR classifier on 3 × 3 image tiles, and navigation is DVL-aided INS with USBL. Results are reported as inspection efficiency and distance efficiency.
The paper
"Autonomous Subsea Cable Search and Tracking with Graph-Optimised Priors and Visual Tracking"
Ibrahim Fadhil Djauhari, Adrian Bodenmann, Samuel Simmons, Cailei Liang, David White, Susan Gourvenec, Tom Bennetts, Darryl Newborough, Blair Thornton
University of Southampton, Sonardyne International and the University of Tokyo
IEEE Journal of Oceanic Engineering (2026), doi:10.1109/JOE.2026.3712105
arXiv:2606.23606, June 2026
Funded by the Defence and Security Accelerator (DASA, project ACC2037839). Smarty200 was developed under EPSRC grant EP/V035975/1, and the British Telecom Group provided the test cable.
The paper goes much deeper than this video: the full graph and catenary equations, the simulation study, the maps from all three dives, and the detector's results on every seabed type. If you work on cable inspection or AUV navigation, it's worth reading in full.
This is an independent explainer. I'm not affiliated with or endorsed by the authors, the University of Southampton, Sonardyne International or the University of Tokyo, and the research is all theirs. The animations are my own illustrations of what the paper describes, not footage from the trial; drawings marked schematic are illustrations. The board sketches are mine too. Any mistakes in the explanation are mine.
I'm Bora Celik from Piccard. I explain new research in ocean science and robotics.