DARPA’s Subterranean Challenge Scores

The U.S. Defense Advanced Research Projects Agency (DARPA) has already awarded more than $3 million in funding and prizes to robot designers competing in its Subterranean Challenge (SubT) underground autonomous robotic vehicle contests.

For the last three years, DARPA has sponsored multiple underground circuit competitions for robotic hardware systems constructed by contestants and tested in real underground environments constructed by DARPA with rock, concrete, steel, wood, and dirt. Concurrently, DARPA has sponsored software robot competitions programmed by contestants and tested in virtual underground facilities constructed by DARPA with software simulations (akin to video games). So far, DARPA's SubT has successfully advanced the state of the art in making quick, accurate maps, as well as demonstrating the need for better cooperative skills among different types of autonomous robots.

"Regarding autonomy, robots have a range of control depending on their environment," said Sanjiv Singh, a consulting professor at Carnegie Mellon University's Robotics Institute and editor-in-chief of the journal Field Robotics. Singh said the journal " covers the areas where robots must operate outside of a controlled environment — like a warehouse — which includes the mining tunnels, urban underground facilities, and the geological caves featured in DARPA's Subterranean Challenge."

Like DARPA's previous Challenges for surface autonomous cars and trucks, the point of the Subterranean Challenge is to pioneer novel new approaches for underground autonomous vehicles, including mapping, navigating, and searching underground environments. DARPA's Subterranean Challenge already helped perfect the technology to quickly make accurate maps of underground environments.

"The 2002 Quecreek Mine rescue near Somerset, PA, for instance, was caused by the inaccuracies in the underground maps in use at that time. The poor-quality maps allowed the miners to break through into an adjacent flooded mine shaft, which in turn flooded the Quecreek Mine, trapping nine miners in an air pocket," said Singh. "DARPA's Subterranean Challenge proved that using robots to map mines can greatly improve the quality and speed with which accurate mining maps can be made."

Another goal of the Subterranean Challenge is to improve the cooperation between different types of robots to achieve a common goal, such as traversing an underground environment with various states of stability that would make it too dangerous for human explorers. Here, however, the Subterranean Challenge demonstrated that more work needs to be done to improve such cooperation. "In subterranean spaces, like the mining tunnels, urban facilities, and geological caves in the DARPA Subterranean Challenge, you need to use rolling, walking and flying robots together," said Singh. "The challenge here is specifically looking for coordination across different types of robots to identify and locate special objects that might be in the environment, which is a big deal given the rugged environment and potentially poor visibility. This need answers the call for improved search and rescue; also, the DHS [Department of Homeland Security] needs to check smuggling tunnels, as well as the military need to perform surveillance operations underground."

According to DARPA program manager Timothy Chung, the advances in robotics created to meet the Subterranean Challenge already are being deployed commercially from "dozens of open-source repositories to which the teams are contributing. We've already seen mention and use of software developed at the SubT Challenge in other contexts, even something as simple as brightening imagery—which may not have been a commonplace piece of software that was necessary for real-time operation of robots, but once you are in low-light settings became a real need. And now that's a utility folks are making use of independently of the SubT Challenge; [in fact,] you are seeing a permeation of those insights and technologies already happening, whether via commercialization channels or through advanced research, which is one of the functions of challenges."

The software robot teams chose from among prefabricated DARPA components to assemble their complements of autonomous robotic vehicles. All the code used to build virtual ground-based robots and other virtual robots, as well as the virtual circuit software itself, is available as open source, including 252 software models and 11 prebuilt software cave circuits. The virtual cave circuits include both open and tightly-constrained passages, rock falls, and vertical "chimneys," as well as large caverns that must be meticulously searched for artifacts. A total of 20 different artifacts were strewn along the cave circuits, each of which needed to be located by the robots. The artifacts included software simulations of fire extinguishers, backpacks, cellphones, helmets, rope, and a survivor (mannequin). The wheeled, tracked, and quadcopter software drones used virtual visual, thermal, and lidar sensors, to navigate and locate artifacts.

Accomplishments in the Subterranean Challenge have come from both hardware robots in real-world environments of rock, concrete, steel, wood, and dirt, as well as from software simulations of virtual underground environments. In the most recent phase (ended Oct. 15, 2020) of DARPA's Subterranean Challenge, software robots competed in eight virtual cave circuit simulations, with the $250,000 first-place award going to the self-funded team Coordinated Robotics; second place was awarded to the Bayesian Adaptive Robot Control System team (BARCS, which was ineligible for prize money since it had accepted DARPA funding), while third place and $150,000 in prize money were awarded to self-funded team Dynamo, and fourth place and $100,000 went to self-funded team Robotika. See more details, including a statement from each team, here.

The recent Cave Challenge was the last of the preliminary contests leading up to the final event next fall in which all teams—including those competing for the first time—will be eligible for prizes in both real-world hardware and virtual-world software competitions.  In this finale, hardware robots in real-world environments, and software robots in virtual-world environments, will concurrently run otherwise identical underground circuits that combine the features of all previous subterranean environments: tunnel, urban, and cave.

The competition offers a $2-million prize to the best hardware robot team, plus $750,000 to the best software robot team, $500,000 to the second-place software robot team, and $250,000 to the third-place software robot team.

May the best robot win.

 

R. Colin Johnson is a Kyoto Prize Fellow who has worked as a technology journalist for two decades.