Underground Challenge

Teams from around the world gathered in a vast, rocky cavern in Louisville, Kentucky, in September 2021. They were there to navigate a complex underground obstacle course they’d never seen before. But humans were forbidden to enter the competition area. The only ones allowed on the course were robots.

The U.S. Defense Advanced Research Projects Agency (DARPA) created the Subterranean (SubT) Challenge to encourage engineers to develop robots that can explore underground environments too risky for humans to enter. For example, after an earthquake or mine collapse, robots could search through rubble to locate survivors. They could also map unknown areas so first responders would know exactly where to go and what dangers awaited them.

To mimic these scenarios, robots participating in the SubT challenge needed to traverse a complicated route on their own, with little or no help from humans. “It was the Super Bowl of autonomous robotics,” says Sean Humbert. He’s a mechanical engineer at the University of Colorado Boulder who led one of the teams that competed in the challenge. The event pushed the limits of robotics technology—which is exactly what organizers wanted.

It was September 2021. Teams from around the world gathered in a vast, rocky cavern in Louisville, Kentucky. They were there to navigate a complex underground obstacle course. They’d never seen it before. But humans were forbidden to enter the competition area. The only ones allowed inside were robots.

The U.S. Defense Advanced Research Projects Agency (DARPA) created the Subterranean (SubT) Challenge. It wanted to encourage engineers to develop robots that can explore underground environments. Some of those places can be too risky for humans to enter. For example, after an earthquake or mine collapse, robots could search through rubble. They might help locate survivors. They could also map unknown areas. That way, first responders would know where to go and what dangers awaited them.

The SubT challenge course mimicked these conditions. Robots in the challenge needed to travel a complicated route on their own. They had little or no help from people. “It was the Super Bowl of autonomous robotics,” says Sean Humbert. He’s a mechanical engineer at the University of Colorado Boulder. He led one of the teams that competed in the challenge. The event pushed the limits of robotics technology. That’s exactly what organizers wanted.

When planning the SubT Challenge, DARPA wanted everything in the course to be as realistic as possible. “We spent a lot of time replicating challenges that we see in real-world underground environments,” says Viktor Orekhov, a robotics engineer who designed the course.

The course consisted of three different settings (see Course Map). A tunnel section featured mine cart rails, mud, and water. A cave section included slippery, tight passages and artificial rock formations like stalactites hanging from the cave ceiling and stalagmites along the ground. Finally, an urban section replicated subway tunnels and underground storage areas.

The designers also created dynamic obstacles that would activate when a robot moved past. A section of ceiling might collapse behind the robot, forcing it to find a different way out of the cave. Or thick smoke might pour into an area, interfering with the robot’s vision.

As a robot navigated the dark, winding course, its goal was to find specific artifacts that represented real-life objects or hazards. Mannequins stood in for survivors. Cell phones and backpacks gave clues that survivors might be nearby. Robots also needed to identify hazardous gas leaks, since they could pose a danger to survivors and human rescuers. Teams would earn a point for each artifact their robots correctly classified and reported. The team with the most points after one hour of searching would win.

When planning the SubT Challenge, DARPA wanted everything to be as realistic as possible. “We spent a lot of time replicating challenges that we see in real-world underground environments,” says Viktor Orekhov. He’s a robotics engineer who designed the course.

The course consisted of three different parts (see Course Map). A tunnel section had mine cart rails, mud, and water. A cave section had slippery, tight passages. It also had artificial rock formations like stalactites hanging from the cave ceiling. Stalagmites stuck up from the ground. An urban section had fake subway tunnels. There were also underground storage areas.

The designers also created obstacles that would do something when a robot moved past. A section of ceiling might collapse behind the robot. Then the bot would have to find a different way out of the cave. Or thick smoke might pour into an area. That would interfere with the robot’s vision.

The course was dark and winding. As a robot moved through, its goal was to find artifacts that represented objects or hazards. Mannequins stood in for survivors. Cell phones and backpacks hinted that survivors might be nearby. Robots also needed to identify hazardous gas leaks. Those could pose a danger to survivors and human rescuers. Teams earned a point for each artifact correctly reported. The team with the most points after one hour would win.

Different bots might be better at navigating certain areas, so teams were allowed to send multiple robots into the course. Some robots rolled on wheels, some glided on tracks, and others walked on four legs. A few teams also used flying drones. A team’s bots would have to coordinate with one another to analyze their surroundings and make decisions. “Anybody can get one of those pieces working very well,” says Humbert, whose team was called Team MARBLE. “But getting everything working well together—that’s the real challenge.”

To prepare, Humbert’s team ran computer simulations of their robots and then traveled to local mines to test the machines’ performance. Afterward, they adjusted the code— or computer instructions—written for the robots to follow. “We did lots of field deployments to iron out all the kinks,” says Humbert.

Different bots might be better at navigating certain areas. Teams were allowed to send multiple robots into the course. Some robots rolled on wheels. Some glided on tracks. Others walked on four legs. A few teams used flying drones. A team’s bots would have to coordinate with one another to analyze the course and make decisions. “Anybody can get one of those pieces working very well,” says Humbert, whose team was called Team MARBLE. “But getting everything working well together—that’s the real challenge.”

To prepare, Humbert’s team ran computer simulations of their robots. Then they traveled to local mines to test the machines’ performance. And then they adjusted their code—the computer instructions the robots follow. “We did lots of field deployments to iron out all the kinks,” he says.

One big challenge was figuring out how the robots would find their way while underground. On the surface, bots can use the global positioning system (GPS). But signals from GPS satellites aren’t reliable underground. To solve that problem, Team MARBLE outfitted their robots with radar and lidar sensors. Radar can help map surroundings by emitting radio waves that reflect off objects, pinpointing their location. Lidar does the same but with beams of light. The robots also had cameras for vision, as well as thermal cameras, which detect heat.

One big challenge was figuring out how the robots would find their way while underground. Aboveground, bots can use the global positioning system (GPS). But signals from GPS satellites aren’t reliable underground. To solve that problem, Team MARBLE outfitted their robots with radar and lidar sensors. Radar can map surroundings. It emits radio waves that reflect off objects. That pinpoints their location. Lidar does something similar. It uses beams of light. The robots also had cameras for vision. And they had thermal cameras, which detect heat.

On the day of the SubT event, the teams ran through the challenge one at a time. Their robots could work autonomously. Or a designated human supervisor could send instructions to help the bots through tough spots in the course. But human help was limited. In some places, communication was purposefully blocked. Most of the time, the robots were on their own.

On the day of the SubT event, the teams took on the challenge. They went one at a time. Their robots could work by themselves. Or a designated person could send instructions to help the bots through tough spots. But human help was limited. In some places, communication was blocked. Most of the time, the robots were on their own.

The four robots of Team MARBLE found their first artifact faster than any other team. But the team suffered a setback when one of their walking robots fell and couldn’t get back up. Ultimately, the team took third place with 18 out of 40 artifacts found. The first place winners, Team CERBERUS—with members from universities in Norway, Switzerland, the United Kingdom, and the United States—located 23 artifacts.

Even though it was a competition, the teams encouraged, supported, and learned from each other during the event. Many are using ideas they picked up from the contest to improve their robots. “Everybody was excited about solving some of these really hard research problems,” says SubT course designer Orekhov. “Now, we’re a community of researchers who are continuing to address these problems together.”

The four robots of Team MARBLE found their first artifact faster than any other team. But they had a setback. One of their walking robots fell and couldn’t get back up. Ultimately, the team took third place. They found 18 out of 40 artifacts. The first-place winners were Team CERBERUS. They had members from universities in Norway, Switzerland, the United Kingdom, and the United States. They located 23 artifacts.

The course was a competition. But the teams encouraged and learned from each other. Many are using ideas from the contest to improve their robots. “Everybody was excited about solving some of these really hard research problems,” says SubT course designer Orekhov. “Now, we’re a community of researchers who are continuing to address these problems together.”