I spot one on a clear September morning, while swimming through the lukewarm waters around the Great Barrier Reef. A striking starfish, the color of cabernet and sporting more than a dozen arms, hugs the side of a gently curving ridge of coral. It looks lovely and dangerous, and not just because I know the long spines covering it are lacquered with venom.

This creature, located a two-hour boat ride from Townsville, off Australia’s northeast coast, is known as the crown-of-thorns starfish. While it’s the first one I have ever seen, it’s far from alone. Since 2010, a plague of them, numbering somewhere in the millions, has been methodically consuming the Great Barrier Reef, representing yet another in a series of existential threats to a coral reef system already wounded by intense hurricanes and weakened by bouts of exceptionally warm waters.

Nor is this the first outbreak. Since 1962, their populations have skyrocketed to “outbreak” status on the reef roughly every 17 years, starting north of Cairns and spreading south in waves, the free-floating larvae carried along by currents. After settling on a reef, they transform into tiny stars, impossible to spot. They don’t start eating coral until reaching at least six months, while still smaller than a dime.

Photo by author.

But from there, they begin to grow and eat in earnest, soon resembling their biblical namesake. The starfish become an army that marches on stomachs extruded from their mouths, digesting the thin layer of living coral beneath them, favoring the quicker-growing corals, the hardest hit by bleaching brought on by rising sea temperatures.

All of which is a way of saying: Clearly, this creature before me needs to die. But the two scuba divers whose task it is to kill crown-of-thorns along this reef are nowhere in sight. Before long, the prickly, peckish villain has disappeared. This is where the killer robots come in.

Two days earlier, I was standing in a brightly lit laboratory in Brisbane, 800 miles down the coast, holding a Samsung tablet and planning a targeted killing. In front of me were five underwater robots, each roughly 2.5 feet long and 1.5 feet wide. They were colored yellow and black, like hornets, and came with a sting to match: each one capable of finding and killing crown-of-thorns starfish without human intervention. “This is a RangerBot,” says Matthew Dunbabin, a cheerful roboticist from the Queensland University of Technology.

The idea for RangerBot began in 2005, when Dunbabin and a student created an algorithm that could recognize the pests. While their system correctly identified the starfish about 67 percent of the time — considered state of the art back then — Dunbabin was stymied by a much more practical challenge: the mechanics of actually killing them.

Photo courtesy of Queensland University of Technology.

In the past, people had tried many ways to kill crown-of-thorns: manually pulling them off reefs for disposal on land; delivering an electric shock; chopping up their bodies (only to have them regenerate); or injecting their bodies and upwards of 20 arms with toxic or acidic chemicals — the method favored by the small number of divers working at the time to cull the starfish populations at popular tourist sites.

Dunbabin’s program could recognize the starfish, but he was far from having the technology to kill them. Luckily, he didn’t need to. In 2014, researchers found that a solution derived of salt and bile induced an extreme and deadly immune reaction in crown-of-thorns, knocking out all but the largest starfish with a single 10-milliliter shot.

“That was the game changer,” Dunbabin said. “When we heard that, we thought, ‘Okay, we can revisit this.’”

He and his colleagues began building RangerBot’s predecessor, a yellow torpedo-like autonomous vehicle called COTSbot (COTS is the acronym for crown-of-thorns starfish). Attached beneath its body was a pneumatic folding arm outfitted with a syringe to inject the bile-salt solution. It carried two liters of solution, enough for 200 injections per trip.

To gather images to train the COTSbot’s detection system (and help figure out the best angles and location for delivering the injections), Dunbabin stuck GoPro cameras on the injection guns of the divers combing the reef for starfish. Dunbabin’s team then fed these and other images — more than 100,000 total — into a deep learning model, which, with help from Dunbabin’s colleague Feras Dayoub, learned to distinguish a many-armed crown-of-thorns starfish from, for example, a branching staghorn coral.

Photo by author.

Developing an autonomous killer robot was not without challenges. Images from the internet, used early in developing the robot vision, were often what Dunbabin called “glamour shots,” perfectly framed and lit, rather than the half-hidden starfish coiled around coral that the robot was more likely to encounter. At the same time, the robotics team wanted to avoid getting equipment stuck in a reef while chasing those hidden critters and decided to focus on starfish they considered relatively accessible — that is, with at least 40 percent of its body showing.

COTSbot developed the ability to identify these starfish with 99.4 percent accuracy and, according to Dunbabin, delivered injections to more than 200 starfish in its two-year robotic lifetime. But at more than four feet long and 66 pounds and costing tens of thousands of dollars, COTSbot — which also required an expert to operate — was a proof of concept, not a final tool. It was time for a complete redesign.

The precise reason behind crown-of-thorns outbreaks is unclear, but several factors seem to be at play. First, the starfish grow fast and reproduce abundantly. A single female can release up to 65 million eggs a year. Second, nutrients washing off sources like sugarcane farms flood coastal waters, providing plentiful food for the phytoplankton that larval starfish consume. Finally, people have overfished the predators that eat these invertebrates, on every level. Under these conditions, crown-of-thorns starfish appear to flourish. The Great Barrier Reef, less so.

However, the Australian government seemed slow to address the booms in coral-eating starfish until recently, as the general outlook for coral became increasingly dire. “The government broad-scale control program started in 2012, so that’s two years after this current outbreak initiated,” says Mary Bonin, assistant director of the crown-of-thorns starfish management program for the Great Barrier Reef Marine Park Authority, the lead agency managing the reef.

Photo by author.

Between 2012 and 2017, the government’s crown-of-thorns control program consisted of a single dedicated vessel and no comprehensive strategy to speak of. It operated in the northern, tourist-heavy region, where trained divers injected and killed starfish by the thousands. The government added a second boat in January 2017 and this year will bring the total to eight.

Inspired by the widespread adoption of flying drones, Dunbabin wondered, “Can we create a drone of the sea?”

Recently, federal agencies and universities in Queensland, the state bordering the Great Barrier Reef, have been creating that much-needed strategy, both for how to guide the control program and how to prioritize the research informing it. Ecologist David Westcott is helping lead those efforts, as he says, “So that we’re not just spending millions of dollars every year killing things.”

A major challenge for the culling program is that divers have to revisit the same reef over and over — something Bonin compares to “weeding a garden” — because crown-of-thorns are often concealed in mazes of rock and coral and are more active at night. “Even when you go to a site with a lot of crown-of-thorns, you might kill every one you see, but you’re not seeing all of them,” says Morgan Pratchett, a reef ecologist studying these starfish. “That’s the big problem.”

Inspired by the widespread adoption of flying drones, Dunbabin wondered, “Can we create a drone of the sea?” In 2016, his team at the university partnered with the Great Barrier Reef Foundation and applied for (and won) AUD$750,000 in funding from Google’s philanthropic arm. Going beyond the idea of a killer robot, they proposed developing a low-cost, easy-to-use Swiss army knife of sorts for reef monitoring and management, something that could administer lethal injections to starfish like COTSbot did, but also handle tasks such as surveying coral, gathering water samples, and mapping the ocean floor.

Two years later, RangerBot was born. This robot, roughly half the size and weight of COTSbot and an eighth of the cost, relies on three systems of robot vision: one for identifying crown-of-thorns, one for navigation, and one for image recording and surveillance. It “sees” using one forward and one downward stereo camera, each flanked by a pair of lights, making nighttime missions now possible. Going from five to six thrusters allows better maneuvering around complexes of delicate coral and more precise positioning over starfish — a major difficulty for COTSbot, which had no lateral control. The system, controlled via tablet, is remarkably easy to use, requiring little by way of formal training. Removable batteries that can power missions up to eight hours long add even more flexibility.

Photo courtesy of Queensland University of Technology.

“We’ll never outcompete a human in terms of their dexterity, their ability to look under corals,” says Dunbabin of RangerBot’s starfish-finding and -killing capabilities, “but what we could do is provide them the tools that can help them upscale.”

RangerBot is still in the trial stage, but this suite of features could make it ideal for surveys, providing quick reconnaissance of where starfish are most concentrated and helping control teams optimize efforts. Experts also foresee sending the robots into deeper or more dangerous waters to avoid risks to divers such as higher pressure, sharks, and crocodiles. Right now, “RangerBot is probably most effective as a surveillance machine rather than as a control machine,” says Westcott, who nevertheless eventually anticipates integrating it into the management program. “It’s going to be a long time before we have thousands of these things out on the reef killing crown-of-thorns starfish.”

Currently, Dunbabin has five robots, with another five in the works. He has received interest for applications as varied as seagrass surveys and mooring inspections. In October, he and a collaborator won approximately $225,000 for a coral restoration project, which will involve adapting RangerBot to seed damaged reefs with coral larvae.

“I thought I was going to get hammered by people for building a killer robot,” Dunbabin admitted. Instead, critics wonder if it’s killer enough.

The marine park authority observed a full RangerBot demonstration in late October, which Dunbabin says will lead to yet more features and new applications in the coming months. “People are under the impression that that thing is out there killing starfish right now,” says Bonin, who looks forward to continuing to learn about its potential. “We’re not quite there yet, but it’s exciting technology.”

The day I spent snorkeling near Townsville, two divers culled only about 70 starfish. But one diver, Warren Haydon, a man who looks happily weathered from five decades of diving, says that similar efforts as recently as January yielded around 700 crown-of-thorns. They would run out of time and chemicals to kill all the starfish, he said. When I saw that reef, about 65 percent of its coral was damaged, mostly by crown-of-thorns.

With corals facing so many threats, recovery isn’t guaranteed, but removing one threat could bolster their chances. “Corals are growing slower,” Pratchett says. “But I’m still hopeful that there will be recovery.”

Still, some question the value of addressing the problem at all. No one pretends the culling program will end the current outbreak, and global warming looms large over the Great Barrier Reef’s future. But several experts say tackling the starfish problem feels like something they can, and should, do. As Pratchett said, “It’s not like if someone’s going to die of cancer, you don’t treat their infection.”

Rohan Kilby, like Haydon, has been culling starfish for two years with a local tour company. He doubts a robot could properly inject starfish in just the right spot and kill it. “I’m very, very skeptical a robot will be able to do that,” Kilby says. Bonin raised the same issue.

“I thought I was going to get hammered by people for building a killer robot,” Dunbabin admitted. Instead, critics wonder if it’s killer enough.

Out on the reef, I saw up close what it takes to kill these starfish, whose native populations will one day collapse on their own and, if nothing changes, start the cycle over again. Floating just above a shelf of pale gold coral, I witnessed Haydon steady himself as he inserted the injection gun twice into a dusky blue and red starfish. The distinctive starfish, more than a foot across, made no obvious reaction, and I watched as Haydon used a metal hook to pull it off the reef and let the creature softly tumble to its death.