Listen to this article
To encourage the development of autonomous systems to explore dangerous environments for first responders and the U.S. military, the Defense Advanced Research Projects Agency has been conducting the DARPA Subterranean Challenge. Last week, the agency announced Coordinated Robotics as the first-place winner in the Cave Circuit Virtual Competition. It was the third of four stages in the challenge, which runs from September 2018 through late 2021.
The DARPA SubT Challenge is intended to test and demonstrate the ability of autonomous robots to perceive, navigate, network, and move in a variety of GPS-denied environments. “We want to inspire and deliver robotics technologies that enable situational awareness across diverse underground environments,” said Dr. Timothy Chung, program manager for the SubT Challenge in DARPA’s Tactical Technology Office.
Coordinated Robotics, a self-funded team led by Kevin Knoedler in Newbury Park, Calif., won a $250,000 prize. Knoedler said he learned to use ground vehicles for the Tunnel Circuit and aerial vehicles for the Urban Circuit. Coordinated Robotics then combined their capabilities for the Cave Circuit. The ground robots shared information and deployed communication nodes, and the drones shared data among themselves and flew through vertical spaces. Endurance, coordination, and using the same software for both virtual and real vehicles were the keys to success, Knoedler said.
Cave Circuit purely virtual
Unlike the first two elements of the DARPA SubT Challenge, which had a hardware Systems Track, the 16 teams ran their mapping and autonomy algorithms entirely in a virtual environment. This was in response to travel restrictions and the need for social distancing during the ongoing COVID-19 pandemic, but it worked so well that organizers said they plan to incorporate some simulation in the final, combined circuit.
The SubT organizers built eight “worlds” ranging from 1.5 to 5.2 km (0.9 to 3.2 mi.) in length in the cloud-based SubT Simulator from 73 “tiles” of cave elements, similar to video game design. The SubT technology repository is open-source, noted Chung.
“This virtual competition is opening the doors on who and where such competitions can occur,” Chung said. “It’s like a fantasy football league — virtual competitors could mix and match robots from across teams. New enhancements for the Cave Circuit included new robot models, dynamic environments, and communications ‘breadcrumbs.'”
“Our small team was very iterative in designing worlds,” said Angela Maio, mechanical engineer at the U.S. Army Research Laboratory and virtual competition lead. “[Factors included] percentage of areas lit, percentages traversable by UGVs [unmanned ground vehicles], size and verticality, and how devious we were in placing artifacts.”
“The robots must communicate with each other to figure out where to explore autonomously,” she said. “Different passages are constrained, plus there are some large caverns to search.”
Competitors configured simulated thermal, visual, and lidar sensors and wheeled or tracked robots or quadcopters, for a total of 17 different platforms and 58 configurations. Each team sent its robots and/or drones to navigate the eight cave courses three times, with the best score counting.
The fully autonomous systems needed to locate 20 “artifacts” hidden in the courses to within five-meter accuracy. These virtual artifacts included backpacks, cellphones, helmets, rope, and a “survivor” mannequin to simulate traces of survivors in a mining accident.
“The challenges involved in navigating unpredictable caves include tight spaces, rockfalls, and lack of visibility — all of which, and more, were incorporated into the virtual caves,” stated Chung. “Beyond pride and prize money, competitors in this and other DARPA Grand Challenges are forging new paths that we are confident will lead to important innovations to help both warfighters and first responders.”
Coordinated Robotics deployed nine virtual robots for the Cave Circuit. The team’s systems participants include the California State University Channel Islands, Oke Onwuka, and Sequoia Middle School.
Coordinated Robotics joins other winners
DARPA-funded BARCS (for Bayesian Adaptive Robot Control System), which includes researchers from Michigan Technological University and the Michigan Tech Research Institute, placed second. Only self-funded teams were eligible for monetary prices in the first three circuits, but all teams are eligible in the final event, said DARPA.
Dynamo, another one-man team led by Hilaraio Tomé in Spain, won third place and $150,000. Robotika, which includes Czech company Robotika International, the Czech University of Life Science, and the Centre for Field Robotics, plus Switzerland-based Cogito Team, placed fourth and received $100,000.
Map Quickly located a cellphone within 5 cm and won “Most Accurate Artifact.” NUS SEDS won the “Reality to Virtual Award” because its software could be most easily used on two Huskies and the 1 QAV500 robot it simulated.
In addition to first place in the Cave Circuit, Coordinated Robotics won “Dynamic Dodger,” successfully avoiding 17 simulated rockfalls.
“The DARPA Subterranean Challenge is creating a community of multidisciplinary teams with a wide variety of expertise,” Chung said. “The best solutions to challenges of navigating the underground will likely be found through combining the ideas of our many talented competitors. We’re already seeing SubT software in other contexts, such as brightening imagery in low-light settings. It’s a utility now that some folks are making use of independently.”
Countdown to the final event begins
DARPA said the final event of its SubT Challenge next autumn will include elements from all three subdomains — tunnel, urban, and cave. It said it expects to run the systems and virtual competitions at the same time.
“Hopefully, we’ll have more marsupials,” said Chung, referring to drone/UGV combinations, which combine a drone’s agility in vertical spaces with a ground robot’s endurance. “Many competitors had considered them up until submission day.”
“It’s admittedly a harder system to think about in terms of autonomy, state management, and knowing when to deploy,” he told The Robot Report.
Both DARPA-funded and self-funded teams will compete for up to $2 million in the Systems Competition and $1.5 million in the Virtual Competition. New teams are still welcome to join, said Chung.