Robots to the Rescue
A headless four-legged mechanical creature hops up a flight of stairs in a University of Pennsylvania building, looking like a mountain goat jumping over boulders. Once at the top, it walks over to an office door and sizes up its target—the elongated silvery door handle that’s just beyond its front legs’ reach. So, it leaps up and hangs on the handle for a moment, unlocking the door. Then it drops to the floor and pushes the door open just like a cat—headbutting its way in.
Called the Ghost Minitaur, this peculiar cross between a goat and a cat was developed byprofessorDaniel E. Koditschek'sPhD students Avik De andGavin Kenneally,at University of Pennsylvania’s General Robotics, Automation, Sensing, and Perception (GRASP) Lab. Robots like Minitaur could prove handy in settings that are too dangerous or difficult for humans,andDe and Kenneally havestarted a company, Ghost Robotics, to build them.
Such robots could navigate burning buildings or earthquake-damaged structures to look for trapped survivors or assess damage. His lab’s other models, RHex and the more advanced X-RHex, walk on springs over uneven terrain, a feature that can be useful in a chemical spill, radioactive leak, or other industrial accidents. They can also travel over unstable collapsible soils or walk to the middle of a disaster site and remain there to monitor it if necessary.
The use of robots in disasters is not a new idea. Around the world, drones have become go-to resources for responders, who use them to assess everything from floods and wildfires to chemical releases. On the sea, robotic watercraft have pulled people to safety from sinking boats.
Yet there remains a crucial need for robots that can work within human-scale environments and perform the type of tasks first responders or canines might undertake while sparing humans and dogs from dangers and health hazards.
They would need to be smarter and more autonomous than existing robots, and agile enough to walk up stairs, climb walls, crawl over rubble, and worm their way under collapsed buildings.
Such robots could also deliver capabilities that people do not have naturally. They could work around the clock without fatigue or loss of precision. They could capture and crunch data faster than any human, reaching decisions faster in situations when every minute can be a lifesaver. And in the worst-case scenarios, they are dispensable—while losing an expensive platform is not an optimal mission outcome, it is better than losing a human life.
Recommended for You: An Army of Tiny Robots to the Rescue
Building such robots remains a challenge. Although the performance of autonomous vehicles—and robots—has improved dramatically, many search-and-rescue platforms would benefit from better dexterity, obstacle avoidance, and better communication and decision-making abilities.
That’s why rescue robots like Minitaur are most often tested in the lab than deployed in the field. And researchers are making progress. Minitaur’s next model, for example, will be able to navigate through a room full of objects. Yet it is still far from the intelligent, dexterous helper that rescue teams would like.
“We are still far away from building anything that they would trust as much as a dog,” Koditschek said. “But we are working on it.”
And so are other teams of roboticists all over the world, each trying different approaches and methods.
Methodologies and Missions
At Ben-Gurion University of the Negev in Israel, David Zarrouk lines up his set of search and rescue robots on his lab floor. Radio-controlled, the robots come to life and swarm around like a motley herd of roaches of varied sizes and shapes.
For Zarrouk, the name of the game is simplicity. That was the lesson learned from the 2015 DARPA robotics challenge, in which humanoid emergency robots did not fare particularly well and some failed miserably.
“The simpler the mechanism, the fewer moving parts there are, the fewer possibilities for failure you have,” Zarrouke said.
His roach robots feature very simple designs but they all roll on a unique set of wheels that look more like a fan with three triangular blades than a circle. Called spoke wheels or rimless wheels, they give the robots insect-like agility, so they can crawl over piles of debris and uneven terrain.
The robots’ structure is also unusual, resembling an “H,” with a central body and two parallel pods. The pods house two spoke wheels each and can pivot about 180 degrees. This lets them position their wheels at different angles or stretch them out flat—a handy feature that enables them to climb up pipes and vertical tunnels.
It also helps them right themselves if they fall and land upside down. They can simply pivot their wheels from above to below their body and continue on their way. As Zarrouk notes, simple mechanisms recover more easily.
Zarrouk takes a bottom-up approach, designing a specific functionality first—such as climbing over rubble or hopping up the stairs—and then tries to find useful applications for it in real-world scenarios.
This is one of the two standard methodologies roboticists take when building any type of robot, explains Robin Murphy, director of Texas A&M’s highly regarded Humanitarian Robotics and AI Laboratory (formerly known as the Center for Robot-Assisted Search and Rescue). The other strategy is top-down design. Both have their pros and cons.
“They are two different methods, but both are totally valid,” Murphy said.
With a top-down method, engineers start with the capabilities first response teams need for a mission, then try to build platforms that meet those requirements. This practical line of attack gives rescuers the immediate help they need, Murphy said.
The bottom-up approach involves fundamental research, and uses prototypes to reveal unexpected pitfalls. For example, RHex’s springs could stir up dust that would interfere with a camera’s feed, Murphy said. Yet fundamental research may also lead to unexpected ideas, like Zarrouk’s self-recovery mechanism.
Still, the top-down approach has its limits, too. Responder teams are pressing for more complex robots that can do more than just gather data, Koditschek said. They may want something that acts like a dog or a squirrel with a fair degree of autonomy and fitness.
“We can’t even begin to build what they need,” Koditschek said. “What we have is too primitive.”
That is why experimenting with various functionalities using a bottom-up approach is essential. Without the fundamental research—a better way to walk over broken buildings, navigate a tunnel, or flip over—it is hard to create new innovative platforms.
No matter which design approach roboticists take, they do not aim to replace what search and rescue teams already do well. Instead, they try to provide what is missing. Murphy cites the snake robot designed by Howie Choset at Carnegie Mellon University’s Robotics Institute, as an example.
Choset built his robot to slither through small holes under collapsed buildings and look for survivors. Dogs can sniff out survivors too, but in some cases, they can be fooled by a scent that’s drifting through a pile of rubble, far from where the actual human is trapped.
“That’s where Howie’s robot comes in really handy,” Murphy said.
Another area where humans need robots’ help is data analysis. Robots and drones equipped with still and video cameras can amass so much data, it can overwhelm disaster response teams that try to make sense of it. Meanwhile, every second is precious.
That’s where machine learning can help. AI can parse this deluge of data and determine what’s different or out of ordinary much faster than any individual human.
“The same dataset that would take humans days to look through—and they would miss things—would take a computer 10 to 90 minutes,” Murphy said.
Teaching AI to parse disaster visuals is challenging because no two disasters are alike, and scenery and imagery can differ dramatically, she explained. Yet researchers have taught computers to scan thousands of images of flood debris in real time. While they may sometimes identify a survivor clinging to a piece of wood, more frequently they determine when something that looks unusual and alert responders to zoom in and take a closer look.
This strategy helped authorities plan their response and save people from drowning after Hurricane Harvey flooded the Houston area in 2017.
Collaboration between humans and robots is key. And in some cases, this collaboration can be best achieved by another innovative alliance by combining both top-down and bottom-up design methodologies.
Its treads whirling slowly, TRADR climbs over a heap of stones and broken wooden crates meant to simulate wreckage. As the robot starts down, its camera whirls side to side, assessing the path. It stops for a moment, then rotates a pair of triangular-shaped flippers forward to probe the uncertain surface in front and help climb up and down.
TRADR, which resembles a miniature tank, might look a bit bulky and slow, but don’t be fooled. In German, its name stands for Long-Term Human-Robot Teaming for Disaster Response. These robots are designed to work in tandem with humans and each other.
Rather than choosing between the bottom-up and top-down approach, the group went for the best of both worlds, said Ivana Kruijff-Korbayová, a senior researcher and project leader in the Language Technology Lab of the German Research Center for Artificial Intelligence.
“We used a user-centric design methodology, which means that we work with end users and ask them what they need,” she said. “We establish user requirements and try to fulfill them.”
The original idea was to use TRADR in industrial accidents, which take place in complex and constantly changing environments.
“Near onset of the incident you can have a lot of fires or smoke that come and go,” she said. “You can have structures that are still collapsing and other variables.”
TRADR robots are particularly helpful where situational parameters change over time because they were built to make multiple sorties over prolonged periods of time. They carry a variety of sensors and use AI to identify specific objects, such as barrels, crates, and plumes of smoke or fires. They can also look for partially visible humans stuck among the debris.
More importantly, they work together. When TRADRs enter an area, they begin mapping it, noting what has collapsed, what remains standing, the location of smoke and fire, and so on. By sharing this information, they can map the entire scene rapidly. As things change over hours or days—debris removal opens one path while a building collapse closes another—they continue to update the map. This enables responders to plan their movements without putting humans in danger.
After putting TRADRs through their paces in several test scenarios, including a train-car collision and several industrial plants, the team deployed them in several disasters.
In the aftermath an earthquake in Amatrice, Italy, the team deployed a TRADR and a drone together to gather structural information about a partially collapsed church that was impossible to enter from the ground. Kruijff-Korbayová’s team also worked with fire brigades from Germany, Italy, Ireland, The Netherlands, and other countries, and attracted interest from firemen all over Europe.
The system can understand simple voice commands such as “search for hazards” or “inspect the barrel.” Currently, it speaks English only, but in the future TRADRs will be able to converse in German and possibly other languages.
“Making it a truly multilingual system is an interesting future option,” Kruijff-Korbayová said. “It could be useful in international missions.”
TRADR is not the only rescue robot to have attracted attention from police, firemen, emergency medical technicians, and other first responders.
Drones are already used to assess floods and wildfires, and to locate survivors while there is still time to save them. They are helping understand what happened when the Fukushima nuclear power plant melted down, and saving humans at sea.
The next generation of rescue robots will go further still, getting closer and closer to the action. They will crawl, climb, and slither amid rubble and flames, giving responders the information they need to confront the world’s most dangerous environments.
As they evolve, these robots will need less and less human guidance. Fully autonomous robots are coming, though models that are as faithful and as capable as a well-trained dog are still years away.
“There is still so much more to do.” Kruijff-Korbayová said. ME
Lina Zeldovich is a freelance writer based in Woodside, N.Y.
Also read: Special Report on Robotics