At the Tour-Ed Mine in Tarentum — ordinarily a spot for tourists who want to check out a coal mine formerly owned by Allegheny Steel — there are, periodically, drones and ground robots exploring the cavernous, maze-like depths.
They’re practicing searching underground environments without human intervention, and the stakes are high: There’s a $2 million government prize on the line.
A group of 15 Carnegie Mellon University researchers are one of six teams being funded by the Defense Advanced Research Projects Agency (DARPA) to develop robotic platforms and software that can traverse dangerous underground sites. The sites might be mines, caves or even urban environments like subway stations.
As with other DARPA-sponsored competitions, the immediate goal is to advance technology that can be used in defense applications. The aim is to improve safety for military and civilian first-responders that operate in underground settings, as in the case of the Thai soccer team trapped in a flooded cave over the summer.
But the grueling competitions usually end up advancing commercial technology, as well.
Matt Travers, a systems scientist at the Robotics Institute that is co-leading the CMU team, said the stuff being worked on for DARPA tends to be around 10-15 years away from hitting the market.
The Defense Advanced Research Projects Agency, which is headquartered in Arlington, Va., did not respond to requests for comment in time for publication. It notes on its website that teams competing in the “Sub-T Challenge” will map, navigate and search in three separate situations over the next three years.
The first leg of the competition, which will task teams with completing actions in an underground tunnel scenario, will take place in fall 2019.
No humans allowed
The Subterranean Challenge is not the first DARPA competition that CMU has competed in.
Back in 2007, a Pittsburgh team won the DARPA Urban Challenge, leading an autonomous Chevy Tahoe named “Boss” to a victory. That, in turn, led to an advance in the field of self-driving car research and propelled some team members to create their own companies, like Aurora Innovation in Lawrenceville and Argo AI in the Strip District.
In 2015, two Carnegie Mellon teams competed in DARPA’s Robotics Challenge, which tasked teams with building semi-autonomous robots that could perform human-like tasks, such as locating and closing a valve on a leaky pipe or opening a door.
One CMU team, which built a humanoid robot called Chimp, placed third. Most of the robots fell over or glitched, requiring human intervention to get them back on track.
Chris Atkeson, a professor and researcher at CMU’s Robotics Institute and Human-Computer Interaction Institute, described the state of search-and-rescue robotics in a research paper following that challenge.
“We believe that robots are useless for disaster recovery if they require humans to rescue the robots, rather than the other way around,” he wrote.
In the Subterranean Challenge, lessons learned from past competitions will be crucial. This time, teams can only employ one human teleoperator — meaning the only real option is for underground tasks to be performed autonomously.
That comes with a novel set of hurdles.
Underground, without a global positioning system (GPS), the robots must find a way to communicate.
Sebastian Scherer, co-lead of the CMU team, explained that the team’s system will probably employ up to eight robots using a WiFi network to “talk” to one another and map their environment.
If that network fails, and one robot goes offline, it will still have the ability to search — based on maps stored onboard the robot itself — but it will not be able to relay information to its metallic teammates.
As for the human operator’s approach if a robot goes offline? That’s classified information for now, Mr. Scherer said. He doesn’t want other teams to snag the plans.
One thing is for sure: As the competition progresses, the team will make changes to its system for each subsequent leg and perhaps borrow ideas from competitors.
Simulation doom
Mr. Travers, Mr. Scherer and their team will rely on a simulation software package developed by DARPA to test their robotic system in the most cost-efficient way possible.
There’s an adage in the computer science world that “simulation is doomed to succeed.”
George Hotz coined that phrase. He was the first hacker to unlock or “jailbreak” the Apple iPhone back in 2007 so he could use the smartphone on other wireless networks outside AT&T, then the device’s exclusive carrier.
The adage is meant to emphasize that while simulations are an inexpensive option to test hardware and avoid physically re-performing trials again and again, the digital world is not an ideal place to prove out a concept.
The world of simulations is far more simplistic than the real world, as evidenced by the experience of the self-driving car industry. Companies building autonomous systems rely on simulations to train software algorithms, teaching the system how to react in a myriad of different situations, but America’s roads bring unforeseen challenges.
Even if CMU’s fleet of aerial and ground robots perform perfectly during a simulated trial of a man-made tunnel or a mine, it’s probably best not to start placing bets that those robots will perform just as expected in real life.
“Our system may work perfectly in simulation and the first time we deploy, we may take the exact same software from the simulation and put it in the robot and it drives right into a wall and you go figure out why,” Mr. Travers said.
The CMU team has already landed $4.5 million in backing from DARPA and an undisclosed amount from Boeing and Near Earth Autonomy, a Point Breeze-based startup developing unmanned flight systems.
Mr. Travers hopes to double that number through investments from the private sector and local foundations.
Generally, he said, the trend in DARPA competitions is that CMU teams must double the money they’ve been gifted by the agency to have enough resources. The team even has a crowdfunding campaign with a $10,000 goal.
As development continues, they still have no idea where the first leg of the challenge will actually be held. It’s a wildcard, but it shouldn’t matter, Mr. Travers said.
The whole point, after all, is for the robots to complete missions on their own, without human helpers.
“We’d like to build a system that’s going to be agnostic to the types of mobility challenges that we’ll face,” he said. “And this is certainly a difficult thing to do.”
Courtney Linder: clinder@post-gazette.com or 412-263-1707. Twitter: @LinderPG.
First Published: November 19, 2018, 12:30 p.m.
