Search-and-rescue workers have finally finished combing through the rubble of dozens of buildings that collapsed in the deadly 19 September earthquake in Mexico. For 3 days, they were joined by two robotic snakes made by roboticist Howie Choset’s research group at Carnegie Mellon University in Pittsburgh, Pennsylvania. The remote-controlled snakes, designed in part for such operations, are 5 centimeters thick and nearly a meter long, with 16 joints that let them move easily through tight spaces. Attached to their heads are lights and cameras, which researchers can access through tethered video screens. Choset’s collaborators Matt Travers, Julian Whitman, and Nico Zevallos-Roberts took the robots to Mexico City 2 days after the quake hit. They didn’t save anyone—this time. But Choset, who spoke with Science about the team’s efforts (and is also a member of the Science Robotics board), hopes the lessons they learned make it possible next time. This interview has been edited for brevity and clarity.
Q: Tell me about your snakebots.
A: They look like snakes, but they’re metal. They can spread through tightly packed spaces and get to locations that people and machinery otherwise cannot access without disturbing the surrounding areas. It’s like minimally invasive surgery. We’ve used them in other deployments, too. We’ve gone to nuclear power plants. We’ve actually done archaeology with them. We were in Egypt until 2 days before the revolution. With my snake robots, I have been working on search and rescue for about 20 years.
There’s a lot of research and development that still has to happen in order for these robots to be ready for prime time. The robots performed very well in Mexico, and given where we are in development, I was very proud of what we’ve done. However, we need to do a lot more. In Mexico, we became more aware of the technical challenges, and we also became more aware of the bureaucratic challenges involved in these kinds of situations.
Q: What were some of the technical things you learned?
A: Let’s start with the stupid little ones. Our control case, the thing that controls the robot—in bright sunlight it doesn’t work as well. There were also mobility challenges. These structured environments are insane. Being able to lower the robot in and look around: That we did very well, but at a certain point we want the robot to move like a real snake does, over this harsh terrain. We need to do a lot more, so we’re that much more respectful of what a real snake can do.
In terms of deployment, we figured out you need one person to drive it, one person to tend to the robot at the point of insertion, someone in the middle to manage the tether, and maybe a fourth person to help the driver interpret the images they’re receiving.
The rescue workers also said we need more sensors on the robot. At the very least microphones. After the microphones, they wanted a gas sensor, because there was a strong concern that they were going into an environment that could explode. You may think: Snap a gas sensor on, and off you go. But there’s a lot of robot architectural decisions made around what is on that robot, so we need to go back and design a more versatile interface.
We also want other means of visualization. The most ideal thing is to build up a 3D map of what you’re seeing as you’re seeing it, and be able to rotate the map around like you’re in the movie The Matrix or playing a video game, and then be able to superimpose gas data and other kinds of sensory information. We’ve developed a sensor head that we didn’t get to use that’s a stereo camera and a laser range finder so you can infer depth information.
If we’d gone 2 days earlier, maybe we could have had a better impact. You really have 48 hours to find people. After that, it’s hard.
Q: After 10 years, your robots aren’t outdated?
A: To be clear, we have advanced the controllers and the theory that drives that robot. For hardware, we did get support from [the Defense Advanced Research Projects Agency] to build the next-generation snake robot. We just didn’t bring it down with us because it hadn’t been tested yet.
Q: What about bureaucratic challenges?
A: It took a local businessman 2 days to Google me and then get the Red Cross to invite us down to Mexico. This should be something more systematic. If we’d gone 2 days earlier, maybe we could have had a better impact. You really have 48 hours to find people. After that, it’s hard. I hate to say it, but if you’re trapped in a pile of rubble and there’s a girder resting on your leg, you’re suffering, and after 2 days of no food and water, you give up hope. It’s a horrible, horrible way to go.
There were also logistical problems, where it became a big turf war over who’s doing what. There were locals saying “Don’t go in my factory, we don’t trust the federal government.” And the Red Cross said, “No, we’re not the federal government, we’re the Red Cross.” And there was a lot of distrust. And I’m sure it’s not a misplaced distrust.
Q: How did you find rubble to search?
A: The Red Cross allocated us an ambulance, so Matt and his team drove around packed in an ambulance with our equipment. The Red Cross would say, “OK, go here, we think there are some people trapped, can you go look?” We’d show up to a location and find whoever is in charge, usually the one with the biggest equipment. We’d unpack and then lay the robots out on a nice big blue tarp so the people there knew what they were getting. After they inspected the robots and understood what they can do, the people in charge said, “OK, go to this hole in the pile of rubble.” Because all they could do was maybe stick their heads in, or use a stick with a mirror on it.
One problem we had was, whoever was in charge, I think their patience was a little bit low and quite understandably so. We got down there 3 or 4 days after the earthquake. At that point, the likelihood of finding life wasn’t that high, so it was sort of like, “Yeah we’ll give you a shot,” and then after an hour or so, “Alright we’re giving up, we’re going to bulldoze the place now.”
Q: Have the snakebots found people on other operations?
A: No, there’s never been a rescue operation where a robot has found people, period. In the Twin Towers attack, [my colleague, Texas A&M University roboticist] Robin Murphy’s robots did find some evidence of body parts.
Q: Did you learn any other lessons in Mexico?
A: We need better algorithms to traverse obstacles. We need better interfaces. And then practice. I know it sounds simple, but going through one of these deepens your understanding.