DARPA Subterranean Challenge: Teams of Robots Compete to Explore Underground Worlds

The DARPA Subterranean Challenge will task teams of robots with autonomous exploration deep beneath the surface of the Earth

Nine teams of humans and robots will make their way to the Edgar Experimental Mine in Idaho Springs, Colo., to participate in an initial exercise for the DARPA Subterranean Challenge
Photo: Colorado School of Mines
Nine teams of humans and robots will make their way to the Edgar Experimental Mine in Idaho Springs, Colo., to participate in an initial exercise for the DARPA Subterranean Challenge.

This weekend, nine teams of robots (and their humans) will make their way to the Edgar Experimental Mine in Idaho Springs, Colo. There, they’ll take part in an initial integration exercise for the DARPA Subterranean Challenge, or SubT. DARPA SubT is a challenge on a similar scale to DARPA’s incredible Robotics Challenge that took place in 2015—a series of competitions based on real-world needs, attracting some of the best roboticists in the world with sophisticated robotic hardware to match.

The integration exercise (which is closed to anyone but the participating teams, we definitely asked) is just the first step in a challenge that will involve both a virtual competition and a competition for physical systems, each with multiple circuits culminating in a final that wraps everything together into one epic event. Some teams will get over US $4 million in DARPA funding, and the prize pool for the finals is up to $2 million. We’ll be following SubT through multiple stages all the way until the final event, which is schedule for August of 2021.

In this article, we’ll get into the details of what SubT is all about, and in a separate post later today we’ll put everything in context with an interview with DARPA program manager Dr. Timothy Chung.

Here’s why DARPA decided to make their next robotics challenge a subterranean one:

The primary scenario of interest for the competition is providing rapid situational awareness to a small team of operators preparing to enter unknown and dynamic subterranean environments. The layout of the environment is unknown, could degrade or change over time (i.e., dynamic terrain), and is too high-risk to send in personnel.

Each team is envisioned to deploy their systems to provide rapid situational awareness through mapping of the unknown environment and localization of artifacts (e.g., survivors, electrical boxes). As the systems explore the environment, these situational awareness updates are provided via reach-back to a base station in as close to real-time as possible. The urgency in completing the course objectives and providing near-real-time situational awareness updates is a consistent focus of the competition.

Something to immediately point out here is that (as you may have noticed) there really isn’t much of an emphasis on “robotics” by DARPA; they consistently refer to “systems.” This is more of a “challenge for robotics” as opposed to a “robotics challenge,” in that DARPA is making it about solving a particular set of problems, rather than solving those problems with robots specifically. We talk about this in our interview with Dr. Chung, but from what we’ve seen and heard so far, most SubT teams will likely be focusing on systems that are primarily, if not exclusively, robotic. 

The SubT Challenge itself will consist of three separate circuits that teams will run one at a time over the next few years, and then final event will be some kind of combination of all three. Each circuit represents a distinct underground environment: tunnel, urban, and cave.

Tunnel: Human-made tunnel systems, like mines

Urban: Urban and municipal underground infrastructure, like subway stations

Cave: Naturally formed networks of tunnels and chambers

Each circuit has unique challenges. For example, the tunnel circuit may have vertical shafts that teams have to deal with. The urban circuit could include stairs. And the cave circuit will probably be the most difficult to traverse, since the terrain will likely be highly randomized.

For each circuit, teams will have to complete a similar set of objectives. The primary mission is to search for, detect, and provide precise georeferenced locations for somewhere between 10 and 30 things that DARPA is calling “artifacts.” These artifacts could include human survivors (mannequins), doors, electric pumps, valves, backpacks, fire extinguishers, radios or cell phones, and more abstract things like gas leaks. A team’s final score is based primarily on how many artifacts they’re able to find, and how quickly, but DARPA expects that teams will also provide continuous, near real-time 3D volumetric map updates as they proceed through the course.

DARPA has emphasized four technology areas for SubT, including autonomy, perception, networking, and mobility. All of these are driven by the environment of SubT, and they encompass a set of technical challenge elements that DARPA will use to more generally assess system performance. In other words, this is what teams are going to have to deal with, and DARPA is going to be very interested in how they manage to do it:

Austere navigation: The challenge courses are expected to include features such as multiple levels, inclines, loops, dead-ends, slip-inducing terrain interfaces, and sharp turns. Such environments with limited visibility, difficult terrain, and/or sparse features can lead to significant localization error and drift over the duration of an extended run.

Degraded sensing: The courses are expected to include elements that range from constrained passages to large openings, lighted areas to complete darkness, and wet to dusty conditions. Perception and proprioceptive sensors will need to reliably operate in these low-light, obscured, and/or scattering environments while having the dynamic range to accommodate such varying conditions. Dust, fog, mist, water, and smoke are within scope of this challenge element. Extreme temperatures, fire, and hazardous materials are not expected to be within scope.

Severe communication: Limited line-of-sight, radio frequency propagation challenges, and effects of varying geology in subterranean environments impose significant impediments to reliable networking and communications links. The physical competition courses as well as the SubT Virtual Testbed environments are designed to include these severe communications constraints to the extent possible. Teams are encouraged to consider innovative approaches to overcome these constraints, including novel combinations of hardware, software, waveforms, protocols, distributed or dispersed concepts, and/or deployment methods.

Terrain obstacles: Systems are required to demonstrate robustness in navigating a range of mobility-stressing terrain features and obstacles. Terrain elements and obstacles may include constrained passages, sharp turns, large drops/climbs, inclines, steps, ladders, and mud, sand, and/or water. The environments may include organic or human-made materials; structured or unstructured clutter; and intact or collapsed structures and debris.

Dynamic terrain: Terrain features and obstacles may also include dynamic elements, which could include, e.g., mobile obstacles, moving walls and barriers, falling debris, and/or other physical changes to the environment that test the agility of the system autonomy to reason, react, and potentially recover from the possibility of a changing map.

Endurance limits: It is expected that successful systems will need to be capable of a team-aggregated endurance of 120 minutes to be mission-relevant. It is expected that each Circuit run will be between 60-90 minutes and each Finals run will be between 60-120 minutes. This aggregate endurance may require novel deployment concepts, energy-aware planning, heterogeneous agents of varying endurance, energy harvesting or transfer technologies, and/or a combination of various approaches to overcome the various challenge elements.

Yeah, that’s going to be quite a challenge. Oof. With all of this stuff going on, you can understand why DARPA suggests “heterogeneous agents,” because this is all likely to be too much for any one system to manage on its own. And by system, we mean robot, of course—we’re expecting to see robots that drive, walk, fly, and perhaps even jump or climb all working together as part of a single team to complete the challenge objectives. 

This brings us to the teams themselves. We’ll have a separate article on this up today, but so far there are nine qualified teams that will be participating in the SubT integration exercise (STIX). To qualify, teams had to describe their technical approach to autonomy (high-level software architecture, human interfaces), perception (sensors, software, degraded sensing approach), networking (hardware, software, radio frequency spectrum), and mobility (number of platforms, types of platforms, fuel sources, safety considerations). They also had to submit a YouTube video, many of which we’ve managed to dig up, showing that their systems would respond to an e-stop, autonomously move over a short course, and autonomously identify at least three artifacts.

There’s also a separate virtual track that teams can compete in without having to invest in any hardware at all. As with the DRC, it’s being run by OSRF/Open Robotics, and so far it looks like 13 teams are signed up. We’ll have more on the virtual track once things start happening later this year. And registration for both tracks is rolling, so it’s not too late to get involved.

For now, the nine teams at STIX will be spending the weekend experimenting on a tunnel course similar (but not identical) to what they can expect for the tunnel circuit competition this August. Teams are getting a chance to experiment on a realistic course, while DARPA gets a chance to make sure that all of their testing infrastructure and whatnot works the way they need it to.

The (tentative) schedule from this point includes the August tunnel circuit competition, followed by the urban circuit competition in February 2020, and the cave circuit competition in August 2020. There will be another integration exercise in February 2021, and the final event will be August 2021. While this first integration exercise wasn’t public, the other events will be, and we’ve been told that DARPA is putting a significant amount of thought into how spectators like us will be able to follow what’s going on, and our guess is that we’ll hear more about that after STIX. We’ll also be following up with some of the individual teams after STIX, to get a sense of how things went, and learn more about how they’re approaching the challenge.

As you might expect, DARPA has provided an enormous amount of information to participating teams. It’s all public, and you can read through everything here.

Or, you can just get caught up on the most interesting bits with these bullet points:

  • Just to be clear about the autonomy: “It is not expected that manual teleoperation of individual systems will be a viable strategy, and teams should expect to provide at most only high-level interactions due to the likelihood of a degraded communications network.”
  • Unlike with the DRC, DARPA is not going to place artificial limits on communications. They don’t need to, they say—the environment will do it for them.
  • DARPA “does not anticipate prohibiting physical links” (like tethers) between robots and a base station outside the course, but they stress that “teams should seriously consider the limitations imposed by large-scale, potentially dynamic, complex environments.”
  • “No manual physical intervention or entry by any (human) team members on the course will be permitted.” But, if your robot falls over, it sounds like DARPA personnel will be be able to perform an intervention on your behalf.
  • “Teams are permitted to make use of dropped components and leave-behind peripherals,” but if DARPA staff can’t easily find them again after your run, you may not get them back.
  • “The order of Circuit Events is not intended to imply that any one subdomain is easier or more difficult than any other.” Just because tunnel is first and cave is last doesn’t mean that cave is any harder, although I’d expect teams to be more capable by the time they get to the last circuit.
  • “DARPA intends to take multiple approaches to limiting the viability of approaches that leverage prior knowledge from earlier runs,” in keeping with the idea that SubT is about rapidly exploring and mapping unknown environments.
  • “DARPA may choose to utilize automated methods of implementing dynamic obstacles.” Fun!

We’ll be keeping you up to date on STIX and SubT in general as much as we can—hopefully, DARPA and the teams will be taking pictures and video and tweeting all weekend. And don’t forget to check out everything we know about the STIX teams so far: That’ll be in a separate post going up later today.

[ DARPA SubT ]

Robotics News

Biweekly newsletter on advances and news in robotics, automation, control systems, interviews with leading roboticists, and more.

About the Automaton blog

IEEE Spectrum’s award-winning robotics blog, featuring news, articles, and videos on robots, humanoids, automation, artificial intelligence, and more.