FATHOM helps an underwater drone track another using sparse light markers

Researchers led by Xiaorui Wang have described FATHOM in an arXiv paper, a control method that lets an autonomous underwater vehicle follow another one using a few active light markers. Such markers stand out in murky water but give sparse views that can drop out through occlusion, reflections or detection noise.

FATHOM combines marker sightings, inertial data and past actions in a learned model of how the tracking will evolve. A diffusion model proposes several thruster trajectories, and the system predicts the outcome of each and picks the best, updating as new observations arrive. In tests on a real AUV, it tracked reliably in normal and difficult motions with markers partly hidden, running in real time on board without motion capture.