When an island’s electricity goes out, traditional methods of locating a break in the underwater power cable can be quite challenging. Previously, this required dragging up the whole line or using remotely operated vehicles (ROVs) to scan it. But what if there was a smarter approach? A project led by the MIT Lincoln Laboratory is investigating this possibility, focusing on designing an autonomous underwater vehicle (AUV) that can map the line and identify the fault’s location.
This pioneering project is part of an internally funded R&D portfolio focusing on autonomous systems, operated by MIT’s Advanced Undersea Systems and Technology Group. Its mission is to optimize maritime missions for the U.S. military, ranging from critical infrastructure inspection and repair to search and rescue activities, harbor entry, and countermine operations. All by effectively combining the strengths of humans and robots.
However, ideal underwater human-robot interaction still poses a significant challenge. “Divers and AUVs typically don’t team up at all underwater,” says principal investigator Madeline Miller. Tasks like underwater infrastructure repair or mines deactivation usually demand human involvement as even ROVs can’t match human finesse. Rough conditions make rapid movement and complex mental calculations difficult for humans, despite their excellent object recognition abilities. Robots, though not agile, have superior processing power, high-speed mobility, and better endurance.
To reconcile human and robot abilities, Miller and her team are developing hardware and algorithms for underwater navigation and perception. Divers often have just a compass for orientation and count their kicks to keep track of distance. Early orientation can be challenging in murky conditions. Robots need to perceive their surroundings accurately to assist the divers, but optical sensors struggle with darkness and water turbidity. Acoustic sensors produce imprecise imagery, only highlighting shapes and shadows.
Despite these challenges, the team ardently pursues solutions for navigation and perception in unfamiliar underwater locations. They have embraced work by the MIT Marine Robotics Group, led by John Leonard, to develop algorithms that promote diver-AUV teaming. These algorithms have been implemented into a relevant AUV and tried out under practical ocean conditions, using a support boat as a diver surrogate before proceeding with actual divers.
Early tests showed that ocean currents demand more sensing capabilities. The vehicle must frequently determine the range to the diver to obtain an effective estimate of positions over time. Real ocean forces, however, complicate optimization.
The team is now working on an AI classifier to process optical and sonar data mid-mission, calling on human input for uncertain classifications. This feedback loop requires an underwater acoustic modem for diver-AUV communication. But underwater communications present challenges due to limited data rates, pushing the team to explore data compression techniques within these constraints.
Extensive testing with divers has taken place in various locations around coastal New England, including the open ocean near Portsmouth, New Hampshire. The team used the University of New Hampshire’s Gulf Surveyor und Gulf Challenger coastal research vessels as diver surrogates. Finally, last summer saw equipment testing with actual human divers at Michigan Technological University’s Great Lakes Research Center.
As this ground-breaking project nears its end, the team hopes to secure external sponsorship to carry on their efforts and transition their work to military or commercial partners. The task at hand is huge, as our modern world relies heavily upon underwater telecommunications and power cables. Protecting these cables from disruptive actions and maintaining U.S. strategic advantage underwater calls for a combination of high-quality artificial intelligence and human capabilities.
If your company is exploring AI automation, visit implementi.ai to uncover potential opportunities. For more in-depth information about this incredible research, you can read the original article on MIT-Nachrichten.
Diese Website verwendet Cookies.