Students at WPI are developing an underwater robot to capture and harvest this invasive species.
If you’ve ever gone reef diving in Florida, the Gulf of Mexico, or the Caribbean, you may have seen the invasive lionfish. Native to Asian Pacific waters, these intruders are known for their stingers, voracious appetite, and prolific reproduction. They eat many of the small fish that native predators eat, thereby disrupting the ecosystem. The only good that comes from the lionfish is they’re good to eat and command high prices at restaurants, if they can be removed from the water.
A team of five undergraduate students in the Robotics Engineering program at Worcester Polytechnic Institute spent their senior year developing an underwater robot that, when completed by future project teams, should be able to catch lionfish and bring them to the surface for harvesting. The team of William Godsey, Brandon Kelly, Joseph Lombardi, Nikolay Uvarov, and Andrey Yuzvik presented their work to the public on April 20, 2018. Here’s how they developed the robot, which will continue with a new team in the 2018-2019 academic year.
Some robots developed for capturing the lionfish will “vacuum” them into a tube or use electricity to stun them. Instead, the WPI lionfish robot harvester spears them with buoyant spears that then float to the surface where a person retrieves the fish. Figure 1 shows the prototype, which has not yet been tested in the ocean. To learn more about the design, EDN spoke with the project team by phone on April 30.
According to Godsey, some robots need a tether to communicate with the surface. “That won’t be reef legal,” he said. “Many local governments will not allow any kind of cable near some reefs, which is why the harvester operates autonomously.”
“At first, we thought about using electricity or some way to grip the fish,” said Yuzvik. Our professors suggested we develop a way to bring the fish to the surface. We designed a buoyant spear that a robot could deploy, which would then bring the fish to the surface.”
As Figure 2 shows, the robot uses a rod to deploy the spear, which is held in place by a magnet to a chamber that’s part of a custom Geneva mechanism. Once deployed, the rod retracts, causing the spear—now embedded in the fish—to detach and float to the surface. The Geneva mechanism indexes by four steps to the spear opposite that last deployed spear. That minimizes imbalance in buoyancy after each the spear deploys, although the robot still compensates for the mismatch.
The students designed the lionfish harvester to use a BlueROV2 from Blue Robotics to propel the robot. For actual testing, the robots will send commands to the BlueROV2, directing it to the intended location after identifying a lionfish. For initial testing, the lionfish harvester has an LCD display controlled by a SparkFun Redboard with an Ethernet shield (Figure 3). The display lets a diver see the commands.
The project team has not yet acquired a BlueROV2, but has used its documentation to generate the necessary commands from the harvester’s Raspberry Pi controller. “We’ve turned the BlueROV2 into an autonomous vehicle,” explained Kelly.
The actual navigation of the robot once it enters the water will be a job for the next team of students. “We have some navigation algorithms for next year’s team,” said Kelley. This prototype is designed to go to the bottom, then move in a spiral pattern until it recognizes a lionfish through a camera and vision algorithms.
Mounted on the front of the robot, two HD low-light cameras from BlueROV connect to the Raspberry Pi controller over a pair of USB cables. Because a Raspberry Pi lacks the processing power to analyze images in real time, the students added an Intel Movidius Neural Compute Stick (NCS). It has a software development kit (SDK) that supports TensorFlow and OpenCV for machine learning. The SDK also supports the Caffe deep-learning framework, which has its own software library with similar computer vision capabilities as TensorFlow. TensorFlow’s object-detection libraries aren’t supported on the NCS right now, so the software switches over to Caffe and Moblienet Single Shot Detector (SSD). That lets the harvester identify the front and side of a lion fish, the parts of the body it needs to hit with a spear. To do that, the vision system applies bounding boxes and can thus locate the desired area. If the system sees the front of the fish, the robot will navigate to the side of the fish. Once it sees the side, it will home in on the location to hit.
The lionfish harvester must get close enough to the target fish to spear it. Once a lionfish is located, the system applies OpenCV2, which has stereo camera libraries, to create a depth map of the area in front of the robot. By applying the X and Y coordinates to the depth map, the robot then moves forward or backward to line up the target. A second check verifies the position of the fish.
Before firing the spear, the system also looks for nearby divers so they don’t get unintentionally hit. The entire calculating process take from 10 ms to 20 ms, which is fast enough to strike the slow-moving fish. The actual time before firing depends on how fast the BlueROV moves, which has not yet been tested, but the team expects that the system will never take longer than 100 ms.
After firing a spear, the robot rearms itself by retracting the rod using a stepper motor. As the rod retracts, a surgical tube (Figure 4) stretches, creating potential energy. Note the gear teeth on the arm in Figure 2. Two solenoids then release the rod, which moves forward, deploying the spear. After deployment, the rod retracts again and the Geneva mechanism indexes four steps to the opposite side. When all spears have been deployed, the robot returns to the surface. In future designs, the robot will return to the surface after a specified amount of time. If it travels too far from the boat, it will reverse course.
Powering the harvester
Anytime you design a portable system, you must take power into account. The lionfish harvester prototype’s electronics run from an 1100 mAh battery. In service, it will draw power from the BlueROV’s battery. Expected usable time underwater is about two hours, plus time to descend and ascend. Expected maximum depth will be limited by the BlueROV – about 100 m – but because some of the prototype’s parts were made by 3d printing, it won’t be able to withstand that depth. Figure 5 contains more information on the project.
While the students did not test the entire robot under water, they did test subsystems in fresh and salt water. “We tested the spears in the WPI swimming pool,” said Yuzvik. For other systems, they used tanks of water and even some water-filled trash cans.
There are still many opportunities for future work. For example, the spears might send an RF signal identifying their locations. Furthermore, some kind of robot-to-robot communication might be needed to keep two or more robots from chasing the same fish. So, while the lionfish robot isn’t ready for prime time, it’s off to a good start.
—Martin Rowe covers test and measurement for EDN and EE Times. Contact him at martin.rowe@AspenCore.com