Student team's submarine charts its own autonomous course
12/8/2017 3:11:21 PM
Looking at the small submarine that is the centerpiece of the Illinois Autonomous Underwater Vehicle team, there isn’t much that says “nautical.”
Dubbed the Enigma, it has no streamlined hull, only a pair of waterproof cameras and two small motors mounted on either side of a boxy acrylic frame, all connected to a bundle of electronics inside a couple of watertight cylinders. The craft is small and light enough for one person to lug around.
And like so many good ideas, Enigma and the team of University of Illinois students that built it started with a pitch, as unlikely and against-the-odds as the little sub might first appear.
“An undergraduate here came to me and said, ‘Look, I’ve been there on somebody else’s team. We need to have a team here. I’ve got some buddies. We want to do it,’” CS @ ILLINOIS Professor David A. Forsyth said.
The undergrad was Computer Engineering student Shubhankar Agarwal, who was part of a similar team at San Diego City College before he transferred to Illinois, and “there” was the annual International Robosub student competition.
Agarwal pitched the idea as a computer-vision problem, Forsyth’s area of expertise. But outside of his work, Forsyth also is a diver. And that paid off.“I got lucky,” Agarwal said, “because he is really interested in underwater stuff in general.”
With backing from Forsyth – financial and logistical, through the use of his home pool -- Agarwal and the team he assembled built the fully autonomous sub for last July’s competition in San Diego, California. They had a strong initial outing, finishing 22nd out 44 teams, and plan to go back in 2018.
Other faculty who have had key roles include Volodymyr Kindratenko, a research fellow at the National Center for Supercomputing Applications and adjunct associate professor in Department of Electrical and Computer Engineering; and Alexander Schwing, an assistant professor of Electrical and Computer Engineering.
In addition to Agarwal, the team consists of CS @ ILLINOIS students Krishna Dusad, Harshit Agarwal, Akshay Mishra, and Aneesh Durg; Computer Engineering students Adrian Brandemuehl and Raimi Shah; Electrical Engineering students Saleh Ahmed and Mohammad Saad; Nuclear Engineering student Wale Adeyinka; and Aerospace Engineering student Tanay Vardhan.
Beyond their initial success, team members learned how to solve an ongoing series of problems, from raising money to figuring out how to submerge thousands of dollars of electronics without getting it wet and how to turn the cameras into the eyes of the self-driving sub.
With a budget of about $3,000, the team competed against other teams with, in many cases, far more resources – financial and otherwise.
“There are very established high-end competitors,” Forsyth said, noting some universities send teams with budgets that approach push six figures. “Cornell trots out about a hundred-grand worth of submarine every year.”
The students said one Indian university paid to ship a much, much larger submarine built by its students.
“You can buy a car with that kind of money in India,” Shubhankar Agarwal said.
But the Illinois team convinced a handful of sponsors to lend a hand.“At career fairs, instead of handing out a resume, we went with a pamphlet about our submarine,” Dusad said.
Forsyth, Microsoft, Viasat Inc., the Department of Electrical and Computer Engineering, and the university’s Student Organization Resource Fee have all provided parts of that $3,000 budget. A Schaumburg-based company called Sparton contributed inertial measurement units, or IMUs, small devices that provide 3D motion tracking. And Nvidia Corporation donated the sub’s computing power - a Jetson TX2 module.
Ronak Shah of Nvidia’s Higher Education Research unit said company sees its help as an investment in the future.
“Autonomous submarines will take us to places we have yet to learn about,” Shah said. “We see major challenges such as biological oceanography, deep-ocean drilling and exploration, underwater inspection, national defense, polar climatology, and so much more being solved by such endeavors.”
The students also learned how to make the most of resources that are available at a major research university, hauling the parts that would wind up becoming their sub to a lab in the Department of Mechanical Science and Engineering for construction and to the Hydrosystems Laboratory in the Department of Civil and Environmental Engineering for some in-water testing. Affordable pool space that’s large and deep enough is tough to come by, Agarwal said.
The team also found support through the CS @ ILLINOIS chapter of the Association for Computing Machinery. ACM provided money, work space and tools.
The problem at the heart of the project was in one sense like any other computer vision problem – the sub needs to see and identify what it sees.
Several members of the team were part of a class Forsyth taught in which they worked with an autonomous car. The technology the car used to navigate is essentially the same as that used by the sub, Dusad said.
But the sub, of course, needs to see under water.
So the dataset is very different than that needed by the car, which relied on an already robust dataset of stop signs, pedestrians, other vehicles and so forth, none of which the sub would encounter.
“They haven’t been trained on things like detecting buoys and gates underwater,” Dusad said. “There’s this thing called transfer learning, where if your network has learned all of these general features for detecting objects and lines, if you train it on these new datasets which are specific to your task, it can learn them all in your task really well.”
The underwater environment also poses another computer-vision challenge.“Under water, things kind of blend together. Everything’s a little blue or green, and so the contrast is much lower. That becomes a problem for traditional computer-vision methods,” Brandemuehl said. “But with neural networks, as long as you give it enough data of images that are taken under water, it learns to capture those properties of the data.”
In a video of its brief competition voyage, Enigma’s small motors push it slowly through an underwater gate, the first task on the course and one that many of the subs failed. But as Enigma headed toward the second challenge, the allowed time expired.
For the 2018 competition, the team plans to build two submarines – “We’re planning to call them Urbana and Champaign,” Agarwal said.
And they’ve found another company that is trying to help, this one closer to home. OceanComm Inc. is a startup in the Research Park at the University of Illinois that works on wireless underwater communication for the oil-and-gas industry.
“We’re going to have communications between two submarines, and no team in the competition has ever had that before,” Agarwal said. “Hopefully next year we’ll have it.”