How Robotics Teams Prepared for DARPA's SubT Challenge

Roboticists share what they've learned so far from DARPA's Subterranean Challenge, and how they readied their bots for the next event—the Tunnel Circuit

6 min read
Image of a DARPA mine
Photo: DARPA

Eleven robotics teams from around the world are in Pittsburgh this week for the Tunnel Circuit event of the DARPA Subterranean Challenge. Our other article is full of details about what the competition will be like, based on what we've heard from DARPA and the videos we were able to find from STIX, the SubT Integration Exercise that took place in a Colorado mine in April. But STIX was a closed event, so to get a sense of how things really went, we needed to talk to some of the SubT systems track teams.

We spoke with a few teams (mostly those where we knew someone who'd be able to answer our questions) to learn about their experiences at STIX, and how they've been preparing for this week’s Tunnel Circuit. The comments below come from:

Team CSIRO Data61 (Nicholas Hudson and Navinda Kottege)

  • Commonwealth Scientific and Industrial Research Organisation (CSIRO), Australia
  • Emesent, Australia
  • Georgia Institute of Technology

Team PLUTO (Camillo J. Taylor)

  • University of Pennsylvania
  • Exyn Technologies
  • Ghost Robotics

Team COSTAR (Joel Burdick)

  • Jet Propulsion Laboratory
  • California Institute of Technology
  • Massachusetts Institute of Technology
  • KAIST, South Korea

Team CERBERUS (Kostas Alexis)

  • University of Nevada, Reno
  • ETH Zurich, Switzerland
  • University of California, Berkeley
  • Sierra Nevada Corporation
  • Flyability, Switzerland

Team Explorer (Sebastian Scherer)

  • Carnegie Mellon University
  • Oregon State University

What was it like for you and your team to participate in STIX?

Team CERBERUS: Participation in STIX was a defining moment for our team as intensive research activities were merged and integrated toward a fully functional system-of-systems approach on subterranean robotics. Team CERBERUS builds on the collaboration of walking and flying robots and associated perception and autonomy algorithms, which we believe have the potential to allow robotic systems to conquer the underground world. However, as always with research, enabling the field-readiness of a novel solution is a challenging—yet exciting—process! Preparing and participating at STIX was all about our preparation as a team and the readiness to perform. 

Team CSIRO Data61: STIX was seven months into the project, and it was really the first time we had all our team (CSIRO, Emesent, and Georgia Tech) testing together. Transportation and logistics were a challenge, not just integrating systems together in this time frame but preparing them for overseas travel as well. It was a great experience, both seeing the scale and expectations from DARPA, and talking to other teams (CERBERUS and PLUTO), sharing experiences as well as things like tarps and battery safety bags. We achieved what we needed to at STIX: localizing and scoring an object against the DARPA servers, flying the drone, and navigating the robots in their environments. 

Gost I like Team CERBERUS’s ANYmal with the headlight eyeballs. Photo: ANYbotics

What were some specific challenges that surprised you at STIX?

Team CERBERUS: During STIX, DARPA showed us a bit of what it takes to be part of such a competition—to (try to) be DARPA ready! We followed a realistic approach and intentionally did not try to push too many uncertain boundaries at STIX. Nevertheless, what we found very challenging, both during our robotic explorations and based on our broader understanding of the Edgar mine, was the combination of a) narrow settings, b) very challenging communications, and c) sensor degradation. Furthermore, the ability to perform as a well-synchronized system-of-systems (and team) within 60 minutes requires absolute readiness and field hardening. STIX gave us a feeling of what level of challenges we will have to deal with and enhanced our excitement and dedication to the project. 

Team PLUTO: One thing that we found interesting was the extent to which even in a mine, which is a human-built environment, the ground surface can be pitted and rutted with train tracks, rubble, and miscellaneous debris, which pose serious trip and snag hazards for ground-based robots.

Team Explorer: One thing was dust, especially for flying. Some mines are relatively wet, but the mine in Colorado was quite dry, so we had some big dust clouds when we were flying, and that confused some of the lidars a bit. That's something that we had to accommodate. Some other thingsthe places where DARPA hid the objects, some of them were very difficult. And then that DARPA put up fog machines was quite surprising.

Team CoSTAR: Because our team had already been testing in local mines (not nearly as big as the STIX or Tunnel Circuit mines), we knew several things to expect. Wireless communication is really tough in caves. We knew that, but we were really taken aback by how many problems we and other teams had with this issue. Moreover, DARPA really showed up with their A-game for STIX—they put some of the scoring items in tough places. The tunnel was also more complicated, in terms of its branching structure, than we had prepared for, and we weren't ready as a team for the vast size of the Edgar mine.

Team CSIRO Data61: There were narrow tunnels, many unstructured spaces, and one part of the tunnel system was the dustiest mine our drone had been in. There was also a smoke machine which made it difficult and disorientating for even people to walk through. The aggressive terrain pushed our UGV platform reliability. Being expected to find a cellphone a few meters under a metal grate really highlighted the expectations from DARPA.

NCTU Team NCTU is bringing a blimp! Photo: NCTU

How have you been preparing for the SubT Tunnel Event?

Team CoSTAR: The main preparations for SubT? Lots of long hours by core team members. Lots of testing and “gaming,” as we call it (setting up mock competitions). Many hackathons to effect subsystem integration. And many trips to the Eagle Gold mine in Julian, Calif. (still a 3-hour drive from us) for testing and shakeout.

Team PLUTO: In preparation for the tunnel circuit in August, our team has been testing in a mine facility in Lansford, Penn., around 2 hours from our home base in Philadelphia. This mine has been a really interesting test site for us since it was continuously operated between 1855 and 1972, so the tunnels and shafts reflect the history of mining in the area. Because of its age and the nature of the mining operation, this facility has several unusual features, such as tight, slanted passageways and drainage ditches that pose unique challenges.

Team Explorer: The last couple weeks, we've been testing a lot at local mines to iron out all the issues and get the system robust. Some of the things we're explicitly looking at is how to cope with the fog and the dust, and we've spent a lot of time on that to make sure we can successfully navigate through those conditions. Our strategy for communications has been to lay out a network of mesh Wi-Fi as the robots go along, but the robots can also go out of communication and then come back and report what they found.

Team CSIRO Data61: The first thing we did after STIX was work with BIA5 (a local Brisbane company) to procure a Titan platform, which is far more robust than the tracked platform we were using. We also completed building and qualified our hexapod platform, rebuilt our common sensing payload to be IP rated, and created a system to drop mesh network communication nodes.

The whole team (CSIRO, Georgia Tech, and Emesent) has also put considerable effort into autonomous exploration—finding 3D “frontier points” in a shared map, and beyond-communication-range behaviors to navigate to these frontiers and then return to the communications backhaul network. All the platforms are capable of onboard object detection, and we have focused on verifying the end-to-end system accuracy: from the pose of the object with respect to the robot, to the robot’s localization using multiagent SLAM, exploiting tunnel loop closures.

Team CERBERUS: In preparation for the Tunnel Circuit, we have a dual approach on a) fundamental research to solve the broad and general challenges of subterranean robotic autonomy, while simultaneously b) maximizing our competition readiness given the current state of our technology. This also implies multiple underground field deployments (in mines and tunnels), system integration activities across our teams, developing of specific modules related to competition needs and challenges (e.g., automated artifact detection and reporting), and of course, lots of team training. Last but not least: It is also a lot about the spirit of the team, the fun of the research, and the excitement of the competition. We are thrilled to be members of this effort and the collective journey of the robotics community!

Ghost Team PLUTO has a whole, um, herd of Ghost Robotics' Vision 60 quadrupeds. Photo: Ghost Robotics

There are a bunch of other teams involved in the systems track as well, and we're looking forward to meeting all of them in person. The best way to keep up with the competition is probably to follow the hashtag #SubTChallenge on Twitter.

[ DARPA Subterranean Challenge ]

The Conversation (0)

How the U.S. Army Is Turning Robots Into Team Players

Engineers battle the limits of deep learning for battlefield bots

11 min read
Robot with threads near a fallen branch

RoMan, the Army Research Laboratory's robotic manipulator, considers the best way to grasp and move a tree branch at the Adelphi Laboratory Center, in Maryland.

Evan Ackerman

This article is part of our special report on AI, “The Great AI Reckoning.

"I should probably not be standing this close," I think to myself, as the robot slowly approaches a large tree branch on the floor in front of me. It's not the size of the branch that makes me nervous—it's that the robot is operating autonomously, and that while I know what it's supposed to do, I'm not entirely sure what it will do. If everything works the way the roboticists at the U.S. Army Research Laboratory (ARL) in Adelphi, Md., expect, the robot will identify the branch, grasp it, and drag it out of the way. These folks know what they're doing, but I've spent enough time around robots that I take a small step backwards anyway.

The robot, named RoMan, for Robotic Manipulator, is about the size of a large lawn mower, with a tracked base that helps it handle most kinds of terrain. At the front, it has a squat torso equipped with cameras and depth sensors, as well as a pair of arms that were harvested from a prototype disaster-response robot originally developed at NASA's Jet Propulsion Laboratory for a DARPA robotics competition. RoMan's job today is roadway clearing, a multistep task that ARL wants the robot to complete as autonomously as possible. Instead of instructing the robot to grasp specific objects in specific ways and move them to specific places, the operators tell RoMan to "go clear a path." It's then up to the robot to make all the decisions necessary to achieve that objective.

Keep Reading ↓ Show less