You wake up underground. You’re dehydrated, your head’s throbbing, and there’s a deep gash on your forehead that’s bleeding heavily. Maybe you’re stuck in a cave system, trapped in a narrow tunnel, arms by your side. Perhaps you’re wedged in a storm drain that’s slowly filling up with water. Or possibly it’s a mine shaft where the power has gone out, plunging you into terrifying pitch blackness. Then you hear something. It’s only faint, but you know what it means: Help is on the way. Only it’s not from a human rescue team. From the rumbling sounds in the distance, it seems that the search-and-rescue team is sending in the robots. Immediately your sense of relief gives way to a feeling of trepidation. With time running out, and maybe only one chance to get this right, you pray that the robot they’ve picked is up to the job.
This nightmare of a scenario is one that, hopefully, will never befall you. But it’s one that DARPA, the United States’ Defense Advanced Research Projects Agency, is desperately trying to figure out an answer to. And they’ve got $2 million earmarked for whoever can help them.
To work out just what the perfect underground rescue robot should look like, DARPA has set up a competition, the latest in its series of Grand Challenges. Called the Subterranean (or “SubT”) Challenge, this contest — which runs through 2021 — aims to uncover the best the robotics world has to offer in the way of rescue bots. The competition is open to everyone from established robotics researchers to what DARPA Project Manager Dr. Timothy Chung refers to as self-funded “tinkerers” from around the world. All you need to be in with a shot of scooping up the seven-figure prize is to have created a robot that’s able to map, navigate, and search a variety of complex underground environments during time-sensitive combat operations or disaster response scenarios. Beyond that, there are no fixed guidelines about what these robots should look like.
The call for entrants has resulted in a massive groundswell of interest and entries. These range from walking quadruped robots like the four-legged ANYmal robot Digital Trends has covered at length to flying robots which use lidar, the bounced laser technology that helps self-driving cars “see.” Recently, the creations of 11 of the top international robotics teams went underground to be put to the test in the most challenging of environments.
Solving the big problems
For an organization whose name invokes images of shadowy government secrecy, DARPA’s Grand Challenges are surprisingly public. And that’s exactly the point. Since 2004, DARPA has staged similar contests every year or so, with the aim of giving creators around the world a nudge, both monetarily and inspirationally, in the direction they hope that technology will develop in. The prizes on offer are, in essence, sponsorship deals for high-payoff research bridging the divide between fundamental research and tools for potential military applications.
The Grand Challenges can appear zany; much like DARPA’s work into other oddball research topics like self-guiding bullets and cyborg insect spies. But they often only seem this way because the area they’re exploring is so new. In 2004, for instance, DARPA promised a prize of $1 million for anyone who could build a car that could drive itself on a 142-mile route through the Mojave Desert. The “winning” team made it less than eight miles in several hours before catching fire and shuddering to a halt. That same year, the MIT and Harvard economists Frank Levy and Richard Murnane used self-driving cars as an example of a task that machines were unlikely to ever master due to its complexity. A decade-and-a-half later, we know different. DARPA’s Grand Challenge helped lay out a vision for autonomous vehicles, which is now bearing fruit.
“One of the things that DARPA is always interested in is trying to identify breakthrough innovations,” Chung explained. “Sometimes that occurs outside of the traditional pathways, and we need to look for [ways to open] up the aperture where those innovations can arise. These Grand Challenges are really fantastic for being able to pose a very bold problem and then opening it up to the world. It’s a way of both creating excitement [and also inspiring] those that may not traditionally come out to put forward [possible solutions].”
The SubT Challenge sounds similarly audacious. Right now, the idea of dispatching a robot into a disaster zone, in place of a flesh-and-blood first responder, may strike you as unlikely. I’ve lost track of the number of times that robotics researchers describe “search-and-rescue” applications as the eventual justification for boundary-pushing research with no immediate applicability.
But there is good reason to believe that robots can be used in this way. In recent years, a laser-shooting snake robot has been used to help decommission a nuclear power plant in Europe, the U.S. Army has sought out 3,000 battlefield-ready scorpion robots for bomb disposal, and drones and a robot called Colossus were called into service to help battle a massive blaze at the Notre Dame Cathedral in Paris. In all of these cases, robots were the first course of action. As much as people fret about machines stealing human jobs, each of these represents a situation in which humans are risking their safety, or even lives, by going into dangerous scenarios.
“There’s information they can provide and gather, without humans having to be placed at risk.”
“We don’t want to send in robots just for robot’s sake,” Chung said. “Rather, there’s information they can provide and gather, without humans having to be placed at risk.” This is where he sees the SubT Challenge really contributing.
“Fundamentally we are interested in helping in those situations where, if there’s any way we can reduce exposure of people to risks in these hazardous environments, there’s a contribution to be made,” he continued. “In search-and-rescue scenarios, even if it’s simply keeping people out while there’s information that needs to be accumulated, [robots can be sent in] to generate that first look: where there are hazards, where there are pockets of fresh air, where there is structural instability. All that information helps human responders to respond more effectively.”
The SubT Challenge just completed its first phase. Taking place over a four-day stretch in August in a defunct mine system in Pittsburgh, Pennsylvania, DARPA created a simulated disaster scenario to put the robot entrants through their paces. Chung described it as an “audacious underground scavenger hunt” in which the robots had to journey through a large mine system, populated by objects such as thermal mannequins representing survivors. The robots which participated included 20 unmanned aerial vehicles, 64 ground robots, and one autonomous blimp robot called Duckiefloat.
“The idea is that teams of robots must traverse, overcome, sometimes identify new paths, to go and find those artifacts,” he said. “For every artifact that’s found, they score a point. The teams which score the most points within the allotted time are the competition winners.”
It was a tricky challenge. “The robot teams started out at one entrance, and that’s really all the information they had about the entirety of the mine,” he explained. “These were unknown to them. We didn’t provide them any prior information or maps.”
The winning group on this occasion was Team Explorer, a collaboration between Carnegie Mellon and Oregon State University. The team’s robots were able to uncover 25 out of 40 artifacts in harsh conditions including plenty of water and mud. Team Explorer designed and built two ground vehicles and two drones created specifically to operate in mines. Another group that performed well was Team CoSTAR, a powerhouse crew comprised of NASA’s Jet Propulsion Laboratory, MIT, Caltech, and various others. In the end, CoSTAR’s robots were able to uncover 11 of the 40 artifacts: not exactly a clean sweep, but not a bad showing either.
This is far from the end of the competition. In February 2020, the next SubT Challenge will take place in a so-called Urban Circuit. “You can imagine that being anything from a metro-type transit station to infrastructure like storm drains or sewers,” Chung said. “These are all types of environments that are underground urban settings where an emergency might occur.” (This Urban Circuit is what DARPA recently put out a tweet for, requesting access to “human-made underground environment spanning several city blocks” that include a “complex layout and multiple stories, including atriums, tunnels, and stairwells.”)
After that will be a Cave Circuit in August 2020, in which participating robots must contend with the unpredictability of natural underground environments. The competition is set to conclude in August 2021, with a “Final Event” that incorporates all three subdomains in one giant track. The winner will then be awarded the $2 million prize money.
“One of the fun parts about the SubT Challenge is just because one solution works well in one circuit doesn’t necessarily mean that it will do well in future circuits,” Chung said. “That’s because the diversity of the environments lend themselves to potentially quite different solutions. For example, on the Urban Circuit, we can anticipate much more verticality, where just being a ground robot may not suffice. In the Cave Circuit, [meanwhile], there will be a lot of irregularities due to mother nature’s creativity.”
Does it sound fun? You bet! Is it a massive technical challenge to overcome? Absolutely! But could the valuable research coming out of this project one day be used to save people’s lives? That’s what everyone’s counting on.
- Want to lift 200 pounds without breaking a sweat? Strap into this exosuit
- MIT is teaching self-driving cars how to psychoanalyze humans on the road
- Meet OffWorld, the startup that wants to mine the moon with a swarm of robots
- Team machines: Why the next revolution in robotics is collaboration
- Neuro-symbolic A.I. is the future of artificial intelligence. Here’s how it works