When a Manhattan parking garage collapsed in April this year, rescuers were reluctant to stay in the damaged building, fearing further danger. So they used a combination of flying drones and a doglike walking robot to inspect the damage, look for survivors and make sure the site was safe for human rescuers to return.
Despite the robot dog falling over onto its side while walking over a pile of rubble — a moment that became internet-famous — New York Mayor Eric Adams called the robots a success, saying they had ensured there were no overlooked survivors while helping keep human rescuers safe.
Soon, rescuers may be able to call on a much more sophisticated robotic search-and-rescue response. Researchers are developing teams of flying, walking, and rolling robots that can cooperate to explore areas that no one robot can navigate on its own. And they are giving robots the ability to communicate with one another and make many of their own decisions independent of their human controller.
Such teams of robots could be useful in other challenging environments like caves or mines where it can be difficult for rescuers to find and reach survivors. In cities, collapsed buildings and underground sites such as subways or utility tunnels often have hazardous areas where human rescuers can’t be sure of the dangers.
Operating in such places has proved difficult for robots. “You have mud, rock, rubble, constrained passages, large open areas … Just the range and complexity of these environments present a lot of mobility challenges for robots,” says Viktor Orekhov, a roboticist and a technical advisor to the Defense Advanced Research Projects Agency (DARPA), which has been funding research into the field.
Underground spaces are also dark and can be full of dust or smoke if they are the site of a recent disaster. Even worse, the rock and rubble can block radio signals, so robots tend to lose contact with their human controller the farther they go.
Despite these difficulties, roboticists have made progress, says Orekhov, who coauthored an overview of their efforts in the 2023 Annual Review of Control, Robotics, and Autonomous Systems.
One promising strategy is to use a mix of robots, with some combination of treads, wheels, rotors, and legs, to navigate the different spaces. Each type of robot has its own unique set of strengths and weaknesses. Wheeled or treaded robots can carry heavy payloads, and they have big batteries that allow them to operate for a long time. Walking robots can climb stairs or tiptoe over loose rubble. And flying robots are good at mapping out big spaces quickly.
There are also robots that carry other robots. Flying robots tend to have relatively short battery lives, so rescuers can call on “marsupials” — wheeled, treaded, or legged robots that carry the flying robots deep into the area to be explored, releasing them when there is a big space that needs to be mapped.
A team of robots also allows for different instruments to be used. Some robots might carry lights, others radar, sonar, or thermal imaging tools. This diversity allows different robots to see under varied conditions of light or dust. All of the robots, working together, provide the humans that deploy them with a constantly growing map of the space they are working in.
Although teams of robots are good for overall mobility, they present a new problem. A human controller can have difficulty coordinating such a team, especially in underground environments, where thick walls block out radio signals.
One solution is to make sure the robots can communicate with one another. That allows a robot that’s gone deeper and lost radio contact with the surface to potentially relay messages through other robots that are still in touch. Robots could also extend the communications range by dropping portable radio relays, sometimes called “bread crumbs,” while on the move, making it easier to stay in contact with the controller and other robots.
Even when communication is maintained, though, the demands of operating several robots at once can overwhelm a single person. To solve that problem, researchers are working on giving the robots autonomy to cooperate with one another.
In 2017, DARPA funded a multiyear challenge to develop technologies for robots deployed underground. Participants, including engineers working at universities and technology companies, had to map and search a complex subterranean space as quickly and efficiently as possible.
The teams that performed best at this task were those who gave the robots some autonomy, says Orekhov. When robots lost touch with one another and their human operator, they could explore on their own for a certain amount of time, then return to radio range and communicate what they had found.
One team from Australia’s Commonwealth Scientific and Industrial Research Organisation (CSIRO) took this further by designing its robots to make decisions cooperatively, says Navinda Kottege, a CSIRO roboticist who led the effort. The robots themselves decided which tasks to undertake — whether to map this room, explore that corridor or drop a communications node in a particular spot.
The robots also decided how to split up the work most effectively. If a rolling robot spotted a corridor that was too narrow to enter, a smaller walking robot could come and take over the job. If one robot needed to upload information to the base station, it might transmit it to a robot that was nearer to the entrance and ask that robot to walk back to within communications range.
“There were some very interesting emergent behaviors. You could see robots swapping tasks amongst themselves based on some of those factors,” Kottege says.
In fact, the human operator can become a weak link. In one effort, a CSIRO robot wouldn’t enter a corridor, even though an unexplored area lay beyond it. The human operator took over and steered the robot through — but it turned out that the corridor had an incline that was too steep for the robot to manage. The robot knew that, but the human didn’t.
“So it did a backflip, and it ended up crushing the drone on its back in the process,” Kottege says.
To correct the problem, the team built a control system that lets the human operator decide on the overall strategy, such as which parts of the course to prioritize, and then trusts the robots to make the on-the-ground decisions about how to get it done. “The human support could kind of mark out an area in the map and say, ‘This is a high priority area; you need to go and look in that area,’” Kottege says. “This was very different than them picking up a joystick and trying to control the robots.”
This autonomous team concept broke new ground in robotics, says Kostas Alexis, a roboticist at the Norwegian University of Science and Technology whose team ultimately won the challenge. “The idea that you can do this completely autonomously, with a single human controlling the team of robots, just providing some high-level commands here and there … it had not been done before.”
There are still problems to overcome, Orekhov notes. During the competition, for example, many robots broke down or got stuck and needed to be hauled off the course when the competition was over. After just an hour, most teams had only one or two functioning robots left.
But as robots become better, teams of them may one day be able to go into a hazardous disaster site, locate survivors and report back to their human operators with a minimum of supervision.
“There’s definitely lots more work that can and needs to be done,” Orekhov says. “But at the same time, we’ve seen the ability of the teams advance so rapidly that even now, with their current capabilities, they’re able to make a significant difference in real-life environments.”
This article originally appeared in Knowable Magazine, an independent journalistic endeavor from Annual Reviews. Sign up for the newsletter.
