Students at Worcester Polytechnic Institute (WPI) are helping to combat a major environmental threat by designing an autonomous underwater robot that can hunt for invasive lionfish. Using a combination of floating spears and machine learning, the robot is able to track down and harvest lionfish safely without the need for a human operator.

The lionfish is a major environmental problem. According to NOAA, over a million of the predators infest the warm waters of the Caribbean, Bahamas, and areas of the US south and east coasts. A native of the Indo-Pacific region, the lionfish was unknown in American waters until about 25 years ago when they began to appear and spread in large numbers. DNA analysis indicates that all the lionfish in the Americas from New England to Venezuela are descended from only a few individuals. This suggests that the fish were likely the result of a few private collectors dumping their aquaria into the sea.

The problem with the lionfish is that it not only has a voracious appetite, can lay 30,000 fish every five days, and has no natural enemies in its invaded waters, but it's also so foreign that the native fish simply ignore the it until it eats them. The lionfish even has an aggravating tendency to avoid traps. This means that one of the few effective ways yet found to exterminate the lionfish is to hunt it down using scuba divers.

In a way, that's not a bad idea. The US and state governments encourage lionfish hunting and despite its poisonous barbs, the fish is actually delicious, commands a market price of up to US$20 a pound, and is featured in many upmarket restaurants. The snag is that there are only so many recreational scuba divers, they can only operate down to about 100 ft (33 m) and can't reach many areas where the lionfish can find refuge.

For these reasons, a number of organizations have been looking at robots as a way to more effectively hunt lionfish, but these have tended to be tethered remotely piloted vehicles that need a human operator in the loop to do the actual hunting and killing. So, as part of their Major Qualifying Project (MQP), WPI student groups are working on robotic systems that can not only kill lionfish, but can also identify what is a lionfish and what isn't.

The WPI robot is designed to work untethered and autonomously hunt down, intercept, and kill lionfish without an operator, then send them to the surface for collection.

At the heart of the robot are two systems. One is an AI platform that uses machine learning, advanced computer vision libraries, neural network software, and computer vision models to help the robot identify lionfish. This system uses a database created by looking at thousands of lionfish images as well as images of what isn't a lionfish and shouldn't be hunted, like a scuba diver.

The second part is a type of revolving spear holder. This consists of a rotating cylinder containing eight buoyant spears. When closing in on a lionfish, a metal shaft thrusts the spear into the fish. As the shaft retracts, the spear tip detaches and its buoyancy pulls to lionfish to the surface for collection. Meanwhile, a watertight, air-filled chamber displaces water equal to the spear to maintain the robot's trim.

WPI says the new robot is designed to be compatible with commercial autonomous robots, and that a second MQP team will spend the 2018-19 academic year working on a navigation system that will allow the robot to autonomously set up and carry out a 3D search grid.

"The goal is to be able to toss the robot over the side of a boat and have it go down to the reef, plot out a course, and begin its search," says Craig Putnam, senior instructor in computer science and associate director of WPI's Robotics Engineering Program. "It needs to set up a search pattern and fly along the reef, and not run into it, while looking for the lionfish. The idea is that the robots could be part of the environmental solution."

Source: WPI

View gallery - 4 images