Gallery
Description
The Problem
An autonomous underwater vehicle (UUV) must find and stay locked onto a moving target it cannot see. Underwater there is no GPS, no map, and no target position — the vehicle carries only a passive acoustic sensor. That sensor returns no picture of the ocean, just a single faint number, the detection margin (DM), that only whispers whether the target is getting warmer or colder.
That number is not a fixed property of the target: it depends on where the UUV puts itself — its depth (relative to the sound-speed profile), its range to the target, its own self-noise (which rises with speed), and the seabed along the bearing. The vehicle must therefore act — change depth, course, and speed — to make its own sensor work, all while flying blind on one scalar.
The mission has two phases:
- Phase 1 — Acquire (blind dive): the UUV may change depth only, to find the depth band where the acoustics open up and win the first contact (DM >= 0).
- Phase 2 — Hold & sharpen (maneuver): once in contact it may change course, speed, and depth to keep contact, push the detection margin up, and close range — without wasting energy or becoming loud.
A further twist: the target is faster than the UUV (~4 m/s vs a 2.5 m/s cap), so it cannot simply be chased down. Success is about positioning, not brute pursuit: reading how the one scalar responds to each maneuver and exploiting it.
Objective (what we optimize):
- increase time in contact
- increase detection margin
- reduce target range
- reduce maneuver cost
One algorithm must achieve this consistently across four fixed scenarios — two low-frequency (500 Hz) and two high-frequency (15 kHz), each with its own sound-speed profile, bathymetry, and target track.
In short: steer an unseen, faster target using nothing but a single "warmer/colder" signal — first to win contact by depth alone, then to hold and sharpen it by maneuvering.