Finders, keepers: search and rescue robots evolve
Image credit: Dreamstime, EPFL, Alain Herzog, Stanford News/Kurt Hickman, University of Zurich
Although humans and highly trained dogs will always be critical in search-and-rescue operations, robots are being developed – taking inspiration from the natural world – that are helping rescue teams save lives.
Walking over the ruins of the Italian mountain town of Amatrice after the violent earthquake of 2016, Dr Ivana Kruiff-Korbayova spotted a little red shoe. “Hardly anything was left standing,” she remembers. “And around the corner there were grapes and tomatoes ripening in a tiny garden – but was anyone left to harvest them?”
As head of the talking robots group at DFKI, the German Research Centre for Artificial Intelligence, she’d been called at short notice to use her experimental robot technology to help in the wake of the earthquake, which devastated a line of hilltop towns and villages in central Italy, killing 295 people and leaving 4,000 homeless. Amatrice, “the kind of picturesque village where you’d go on holiday”, was worst hit, and half of the town’s mostly historic buildings crumbled. Whole roads through the Italian hills had fallen away, and streets were unrecognisable.
Kruiff-Korbayova and her fellow researchers began working with Italian firefighters a week after the quake struck, “when hope of finding any more survivors had gone”, and emergency services desperately needed to know which structures were safe. Two medieval churches were rickety but still standing amid the rubble, and the team working on the EU-backed TRADR project (for robot-assisted disaster response) travelled from all over Europe to assist. When buildings are damaged, they might collapse in different ways, toppling sideways or pancake-flat – and a detailed recce helps experts make a plan.
Getting in wasn’t easy, but the team’s ground-based robots were able to cross rubble to enter the church out of sight of their operator and roam around the interior, taking images, which the team used to build detailed 3D models of the existing buildings. “They were heavily damaged, cracked walls, collapsed ceiling and dust and rubble everywhere.”
Drones complemented the picture with a live feed of the ground-based robots and of another drone entering the church. “We’d never done this before,” says Kruiff-Korbayova. “We’d used multiple robots simultaneously but never in such tight team collaboration – it was an amazing success.”
Every year natural disasters kill some 90,000 people and affect close to 160 million around the world. After disaster strikes, search and rescue teams know it’s critical to find survivors as soon as possible – the longer the delay, the more lives are lost.
“If you believe you can save lives, you’ll take greater risks,” says private fire consultant Andy Elliott. After a fire, buildings are fragile, water-damaged and prone to collapse. There may be live gas and electricity within, or hazardous materials such as acetylene in confined spaces. “You need accurate risk assessments,” he says. “Damaged buildings are dangerous places.”
Many UK fire services already use drones to track events, says Elliott. Drones relay the impact of natural disasters, too – last August, aerial images showed the devastation of the Bahamas after Hurricane Dorian.
In April 2019, camera-equipped drones relayed real-time details about the flames that gripped Notre Dame Cathedral in Paris – with crucial information on how intense the blaze was and how it was spreading. A remote-controlled robot called Colossus sprayed water within the gothic structure’s interior, sparing firefighters from the risk of tumbling timbers. Moving at a maximum 1m/s, the robot, made by Shark Robotics, carried a camera with all-round high-definition vision and thermal imaging and a motorised water cannon.
Using trial and error learning algorithms, an injured robot can recover autonomously by adapting to broken limbs or motors. Just as an animal would, the injured robot tries out compensatory movements, swiftly settling for one that works. No diagnosis or repair is required; instead the robot learns to cope with whatever damage it has sustained. This is an EU-backed project hosted by the French Institute for Research in Computer Science and Automation (INRIA).
A four-legged robot equipped with sensors to detect its terrain, it was developed by Nasa’sJet Propulsion Laboratory originally for disaster relief but is now being adapted to work in the icy worlds of space.
The robot you can’t crush
Mechanical engineers at UC Berkeley have built a mini bot that scuttles like a cockroach and is as tough too – it can withstand the weight of a human. Robust insect-sized robots such as these could eventually be used in earthquake debris to go where dogs and humans can’t, say researchers behind the design.
Emily (Emergency Integrated Lifesaving Lanyard)
A remote-controlled rescue boat from US firm Hydronalix that can go where it’s too dangerous for lifeguards. Individuals requiring help can use Emily as a float and be towed back to safety. A version equipped with sonar was used to map a lake at the base of Mount Everest to estimate the danger of it collapsing.
Getting robots into a disaster zone to poke around on the ground is no mean feat. When you are a robot, the world is full of obstacles.
Drones are frequently used by emergency services, but struggle to fly in confined spaces, can’t carry much and are limited by battery life. Land robots battle with stairs, rubble and door openings.
“Robot snakes are a way to solve the problem of narrow spaces,” says Dr Emma Rushforth, director of Warwick Mobile Robotics at the University of Warwick, “but they are incredibly complex to create.”
Her robotics students are well acquainted with the hazards of disaster zones, and they’re refining a caterpillar-tracked bot equipped with a robotic arm, with the aim of finding survivors amid the rubble of disaster zones. Their bot can climb kerbs and stairs and travel through mud. Its designers will test its prowess at the RoboCup Rescue League – an international robotics event where search and rescue robots tackle some tough scenarios. “Year on year we try to improve on it, but the difficulties are immense.”
It’s hard to drive something you can’t see; “it’s not like a Christmas toy,” says Rushforth. Operators rely on visuals relayed by the robot, which requires a reasonable signal – and that is not always possible through thick rubble or below ground. “You’re pushing limits of battery life and mechanical ability – simply powering the computing required to do the intelligent autonomous stuff is a challenge,” she says.
While robots have ventured into film within Japan’s ruined Fukushima Daiichi power plant, damaged by the 2011 tsunami, radiation plays havoc with the signal. But pulling a communication cable through a disaster zone is fraught with difficulties too, she says. Advances in batteries, motors and materials will help refine search-and-rescue robots to eventually make them fit for purpose.
However, some of the biggest problems come when robots encounter water. Professor Auke Ijspeert is well aware of this. He’s head of the biorobotics laboratory at Switzerland’s EPFL (École Polytechnique Fédérale de Lausanne) and an expert in computational neuroscience and machine learning.
For eight years, Ijspeert and his colleagues have been building an amphibious robot inspired by the natural world. And after studying the ambling gait and swimming motion of a salamander, his team have created algorithms that mimic nature in robot form, using them to design the ‘pleurobot’ – a fantastically complex segmented creation with “a small micro control in every segment so there’s no single point of failure”, says Ijspeert. “It’s very robust.” And pleurobot walks and swims uncannily like the real thing – although it does require a bespoke drysuit. “Making it waterproof, dustproof and sturdy is a challenge, as is finding the right trade-off between size and weight.”
Ijspeert accompanies Swiss search-and-rescue teams on regular drills at their realistic training grounds, complete with ruined buildings, to try to understand what emergency crew need in the heat of the moment. “We don’t want to replace a rescue team, but complement it when it’s too dangerous for humans or dogs,” he says.
Robots can gather data to build detailed 3D maps, use infrared cameras to search for survivors, even transport links for two-way communication. “Our robots have a payload capability so they could transport water and medicine,” says Ijspeert.
While the pleurobot is more of an academic endeavour, some of Ijspeert’s robotics colleagues at EPFL’s Laboratory of Intelligent Systems are busy developing practical applications.
‘In the future, we want to help a rescue team to be faster, safer and more efficient – potentially with a fleet of robots to explore many things at once.’
As any search-and-rescue professional will tell you, rubble and tight squeezes hamper the search for survivors. Inspired by birds that squeeze through narrow gaps by folding their wings, roboticists at EPFL and Zurich University have built a “foldable” drone – a quadcopter that can change shape to pass through narrow chinks. This allows it to manoeuvre in confined spaces. Four arms, equipped with four propellers, can move independently, and a control system adjusts the thrust as the centre of gravity shifts so the drone remains stable. In the future, the team hopes it will be more adaptable still and equipped with enough autonomy to respond to instructions such as “enter that building, inspect every room, then return”.
Training emergency crew to operate drones is an extra headache for services, and scientists at EPFL are researching the viability of a haptic suit – a ‘fly jacket’ –which would allow an operator to physically ‘fly’ the drone with body movements and wheeling arms, with goggles hooked up to an onboard camera, leaving hands free to pinpoint areas of interest, possibly with the help of data gloves.
Another bright idea to come from the Swiss labs is small drones that can shift objects more than 40 times their weight, using winches and gripping technology inspired by geckos and insect feet. Working together, these micro tugging robots could lasso a door handle to pull open a door. Or a drone inspired by insect wings – built to remain rigid when flying, but able to flex upon collision, so limiting any damage.
Some designs are unashamedly experimental, says Ijspeert, but they’re tackling the important questions, such as how much autonomy to give a fleet of flying robots, what situational awareness they require and how a human best interacts with them, as well as how to cope with a flooded basement or a collapsed floor.
“In the future, we want to help a rescue team to be faster, safer and more efficient – potentially with a fleet of robots to explore many things at once.”
While robots are reaching places humans can’t, they’re no match for rescue dogs yet. “Science has a long way to go to catch up,” says Hampshire Fire and Rescue dog handler Robin Furniss, whose dogs were flown out to work amid the Nepalese and Japanese earthquakes. “Fire brigades have massive technology backing them up, but dogs are still the first preferred call.” His dogs were used to sniff for survivors when the disused Didcot Power station collapsed in 2016, and around fallen cliffs in Dorset in 2012, and many more incidents.
“Dogs can search in minutes what would take a team of firefighters days,” he says. “Their sense of smell is phenomenal – they can detect half a teaspoon of sugar dissolved in an Olympic-size swimming pool.” Scientists are beginning to experiment with using dogs to transport robots into a disaster site – Furniss equipped his dogs with transmitters to feed back live signals, though transmitting through rubble was problematic.
Having seen emergency services in action, Ijspeert and his colleagues are optimistic, but realistic. “In the robotics community we can be quite naïve. And we have to be humble in the face of the amazing things being achieved by the rescue community – we will help them, but we will never replace them.”
Search and rescue
Search-and-rescue robots have been deployed in some form for nearly two decades – they were first used to help search the rubble of the World Trade Center amid the devastation of the 9/11 attacks in 2001. Today, their applications vary, from mapping disaster zones, detecting signs of life such as heartbeats and breathing, and distributing water and food.
Researchers at Stanford have developed a soft, flexible robot inspired by vine tendrils. An air-filled tube fitted with a small camera ‘grows’ as air forces the tube to invert, allowing it to manoeuvre around difficult corners and gaps. Operators can guide it, and in the future it may grow using liquid, which could deliver water to trapped survivors or be used to extinguish fires.
Engineers at the University of Leeds have built a wireless scanner mounted on a drone, which could ‘see’ through collapsed or burning buildings to find individuals who may be trapped. Designed to be used by search-and-rescue teams, the system – a collaboration between UK, US and Chinese academics – can scan deep into a building.
The drone, equipped with both a transmitter and receiver, flies around the outside of the building and uses harmless long-range radio waves that can penetrate concrete half a metre thick. A directed signal acts as a radar, bouncing off objects, and information is fed back and interpreted by software on the ground. The system currently detects survivors who are moving, but its creators want to extend it to identify groups of people at a time.
Finder – or Finding Individuals for Disaster and Emergency Response – is a tool equipped with AI that can detect signs of life below layers of rubble. It’s been used in earthquakes such as the 2015 Nepal earthquake and the aftermath of hurricanes. Originally developed by Nasa’s Jet Propulsion Laboratory and the US Department of Homeland Security, it sends a low-powered microwave signal through rubble, and looks for changes in reflected signals caused by breathing or heartbeats.
Sign up to the E&T News e-mail to get great stories like this delivered to your inbox every day.