For years, the U.S. Navy has employed human divers, equipped with sonar cameras, to search for underwater mines attached to ship hulls. The Navy has also trained dolphins and sea lions to search for bombs on and around vessels.
In the last few years, Navy scientists, along with research institutions around the world, have been engineering resilient robots for minesweeping and other risky underwater missions. The ultimate goal is to design completely autonomous robots that can navigate and map cloudy underwater environments — without any prior knowledge of those environments — and detect mines as small as an iPod.
Now Franz Hover, the Finmeccanica Career Development Associate Professor in the Department of Mechanical Engineering, and graduate student Brendan Englot have designed algorithms that vastly improve such robots’ navigation and feature-detecting capabilities. Using the group’s algorithms, the robot is able to swim around a ship’s hull and view complex structures such as propellers and shafts. The goal is to achieve a resolution fine enough to detect a 10-centimeter mine attached to the side of a ship.
The engineering of such an inspection is a thorny computational problem that Hover and his group have investigated for the last decade. The researchers are coming up with algorithms to program a robot called the Hovering Autonomous Underwater Vehicle (HAUV), originally developed as part of MIT’s Sea Grant program.
Fully viewing a massive structure such as a naval combat vessel — as well as all its small features, including bolts, struts and any small mines — is a tricky planning problem, according to Hover. “It’s not enough to just view it from a safe distance,” Hover says. “The vehicle has to go in and fly through the propellers and the rudders, trying to sweep everything, usually with short-range sensors that have a limited field of view.”
The group approached the challenge in two stages. For the first stage, the researchers programmed the robot to approach the ship’s hull from a safe 10-meter distance, swimming in a square around the structure. The vehicle’s sonar camera emits signals that boomerang back as the robot makes its way around the ship; the researchers process the sonar signals into a grainy point cloud. At such a low resolution, Hover says one can clearly make out a ship’s large propeller, though not an iPod-sized mine.
For the second stage of their approach, the researchers programmed the robot to swim closer to the ship, navigating around the structure based on the mesh model. The idea, Hover says, is for the robot to cover every point in the mesh; in this case, each point is spaced 10 centimeters apart, narrow enough to detect a small mine.
One approach, he says, might be to have the robot sweep over the structure much like one would mow a lawn, one strip at a time — a common technique in robotic inspection. But such rectangular surveys can be tedious and time-consuming. Instead, the researchers came up with a more efficient approach, using optimization algorithms to program the robot to sweep across the structures while taking into account their complicated 3D shapes.
