We know much less about the Earth's oceans than about the surface of the Moon or Mars. The seabed is carved with expansive canyons, towering seamounts, deep trenches and steep cliffs, most of which are considered too dangerous or inaccessible to autonomous underwater vehicles (AUV) for navigation.
But what if the reward for going through these places was worth the risk?
MIT engineers have already developed an algorithm that allows AUVs to assess the risks and potential rewards of exploring an unfamiliar region. For example, if a vehicle in charge of identifying submerged oil leaks approached a steep and rocky trench, the algorithm could assess the level of reward (the probability that an oil infiltration existed near that trench) and the level of risk probability of colliding with a trench). obstacle), if it were to take a path through the trench.
"If we were too conservative with our expensive vehicle, saying that its ability to survive was fundamental above all else, we would find nothing of interest," says Ayton. "But if we understand that there is a trade-off between the reward of what you have collected and the risk or threat of moving toward those dangerous geographies, we can take certain risks when it is worth it."
Ayton says the new algorithm can compute risk versus real-time reward tradeoffs, as a vehicle decides where to exploit next. He and his colleagues in the laboratory of Brian Williams, a professor of aeronautics and astronautics, are implementing this algorithm and others in AUVs, with the vision of deploying fleets of audacious and intelligent robotic explorers for various missions, including looking for offshore oil deposits. , investigating the impact of climate change on coral reefs and exploring extreme environments analogous to Europe, an ice-covered moon of Jupiter that the team expects the vehicles to go through one day.
"If we went to Europe and had a very strong reason to believe that there could be a billion-dollar observation in a cave or crevice, which would justify sending a spacecraft to Europe, then we would absolutely want to risk going into that cave," Ayton says. "But algorithms that do not consider risk will never discover this observation that can change history."
Ayton and Williams, along with Richard Camilli of the Woods Hole Oceanographic Institution, will present their new algorithm at the conference of the Association for the Advancement of Artificial Intelligence this week in Honolulu.
A bold path
The new algorithm of the team is the first to allow "adaptive sampling limited by risk". An adaptive sampling mission is designed, for example, to automatically adapt the path of an AUV, based on new measurements that the vehicle performs when exploring a particular region. Most adaptive sampling missions that consider risk usually do so by finding paths with a concrete and acceptable level of risk. For example, AUVs can be programmed to only trace paths with a collision chance that does not exceed 5%.
But the researchers found that risk accounting alone could severely limit the potential rewards of the mission.
"Before we go on a mission, we want to specify the risk we're willing to take for a certain level of reward," says Ayton. "For example, if a path leads to more hydrothermal vents, we would be willing to take that risk, but if we do not see it, we'll be willing to take less risk."
The team algorithm collects bathymetric data or information about ocean topography, including any adjacent obstacles, along with vehicle dynamics and inertial measurements, to calculate the level of risk for a particular proposed path. The algorithm also leads to all previous measurements that the AUV has taken to calculate the probability that such high reward measures may exist along the proposed path.
If the risk-reward ratio reaches a certain value, as determined by the scientists beforehand, then the AUV goes on along the proposed path, taking more measures that return to the algorithm to help assess the risk and reward of other paths such as the vehicle advances.
The researchers tested their algorithm in a simulation of an AUV mission east of Boston Harbor. They used bathymetric data collected in the region during a previous survey of NOAA, and simulated an AUV exploring at a depth of 15 meters across regions at relatively high temperatures. They observed how the algorithm planned the route of the vehicle under three different scenarios of acceptable risk.
In the scenario with the lowest acceptable risk, meaning that the vehicle should avoid any regions that had a very high chance of collision, the algorithm mapped a conservative path, keeping the vehicle in a safe region that also did not have great rewards. In this case, high temperatures. For scenarios of greater acceptable risk, the algorithm outlined bolder paths that drove a vehicle through a narrow abyss and ultimately into a region of high reward.
The team also ran the algorithm through 10,000 numerical simulations, generating random environments in each simulation to plan a path and found that the algorithm "switches risk from reward intuitively, taking dangerous actions only when justified by reward."
A risky slope
Last December, Ayton, Williams and others spent two weeks on a cruise off the coast of Costa Rica, installing underwater gliders, in which they tested various algorithms, including the newest. Most of the time, the algorithm's path planning agreed with those proposed by several geologists on board who were looking for the best routes to find oil leaks.
Ayton says there was a particular moment when the risk algorithm proved especially useful. An AUV was rising in a precarious depression, or landslide, where the vehicle could not take many risks.
"The algorithm found a method to get us up quickly, being the most valuable," says Ayton. "This led us to a path that, while not helping us discover oil leaks, has helped us to refine our understanding of the environment."
"What was really interesting was watching how machine algorithms began learning after the discoveries of various dives and began to choose sites that we geologists might not have initially chosen," says Lori Summa, geologist and visiting researcher. Woods Hole Oceanographic Institution, which took part in the cruise. "This part of the process is still evolving, but it was exciting to see the algorithms begin to identify the new patterns from large amounts of data and to aggregate this information into an efficient and safe search strategy."
In their long-term view, researchers hope to use these algorithms to help autonomous vehicles explore environments beyond Earth.
"If we went to Europe and were not willing to take risks to preserve a probe, then the probability of finding life would be very, very low," says Ayton. "You have to take a little risk for more reward, which is usually true in life, too."
New Algorithm, Metrics Improve Energy Efficiency of Autonomous Submarine Vehicles
Massachusetts Institute of Technology
This story is republished courtesy of MIT News (web.mit.edu/newsoffice/), a popular site that covers news about MIT research, innovation and teaching.
Engineers program marine robots to take calculated risks (2019, January 30)
recovered on January 30, 2019
This document is subject to copyright. In addition to any fair dealing for private study or research,
may be reproduced without written permission. Content is provided for informational purposes only.