How Robotics Teams Are Solving the Biggest Problem at DARPA’s Subterranean Challenge

Supporting long-range communications inside of a mine is extremely difficult, so DARPA SubT teams are trying out some creative approaches

5 min read
Photo: Evan Ackerman/IEEE Spectrum

When DARPA announced its Subterranean Challenge, the agency framed things with a comprehensive list of “technical challenge elements” that it expected to be particularly, you know, challenging. One of those elements was communication constraints, which DARPA said teams should expect to be “severe.” That may have been an understatement, based on what we're seeing at the SubT Tunnel Circuit—teams have had a lot of trouble consistently talking to their robots on the course.

Even though most teams are emphasizing autonomy as much as possible, they do still have to deal with communication challenges, because finding artifacts won’t earn you any points unless the robot reports its location back to base before time runs out. And if your robot has to come all the way back to base to make its report, it’s going to run out of time after finding just a few artifacts.

There’s no magical solution for this, which is awesome, because teams have come up with all kinds of fantastically creative and unique strategies. We went around and asked them about it when we visited the SubT Tunnel Circuit in Pittsburgh yesterday. 

Some quick context to help you understand the challenge here—robots usually do fine through the first several tens of meters just past the mine entrance. Once they turn the first corner, however, they lose their wireless connection back to the base station almost immediately, because radio waves don’t pass through solid rock. And even if the robots manage to hang on to a signal for a little bit, the twists and turns and sheer distances that the robots must travel (hundreds and hundreds of meters, ideally) means it simply isn’t possible to maintain a direct connection back to base. 

Most teams are adopting a robot-to-robot mesh networking approach to help with this, meaning that the robots themselves can serve as network nodes, and if one robot can communicate back to base, any robot that it can communicate with can also communicate back to base—and so on down as long a chain of robots as you can manage. Robots need to move around to do their jobs, though, and teams only have a limited number of robots, so here are some other strategies that teams are using to keep in contact.


plutoPhoto: Evan Ackerman/IEEE Spectrum

This is about as straightforward as it gets—Team PLUTO mounted massive dipole antennas on the butts of their Ghost Robotics Vision 60 quadrupeds, which in their experience improves communication performance by an order of magnitude. The team is emphasizing autonomy, and like many teams, their robots are designed to operate for extended periods without any communications at all, but they do still need to report back on what they found from time to time.

Team Explorer

explorerPhoto: Evan Ackerman/IEEE Spectrum

Team Explorer, currently leading (by a lot) in artifacts found, uses deployable network nodes (pulled halfway out of a protective casing in the picture above) that its robots can drop when the strength of the network starts getting low. Each robot can carry about 10 nodes in the two racks that you can see, and while the nodes are dropped at the command of the remote operator at the moment, the robots will eventually be able to decide autonomously when they need to plop one down. The team also plans to shrink the nodes down, and to improve the dropping mechanism, which can get jammed by mud.


costarPhoto: Evan Ackerman/IEEE Spectrum

The beefy deployable network node idea was a popular one, and Team CoSTAR (tied for second place in artifacts found at the end of day three) sent in multiple robots carrying nodes in several different configurations. 


cerberusPhoto: Evan Ackerman/IEEE Spectrum

DARPA not-so-subtly suggested that teams may not want to rely on a physical tether between the base station and robots (“teams should seriously consider the limitations [on tethers] imposed by large-scale, potentially dynamic, complex environments,” says the agency), but Team CERBERUS didn’t let that scare them. They’re using a fiber-optic tether and a dedicated communications robot with a massive antenna on it to essentially extend their base station deep into the mine.

The tether isn’t long enough to explore the whole mine, of course, and it has to be carefully managed around corners, but even if the robot just makes it down to the end of the first passage and around the first corner, it’s a massive improvement.


NCTUPhoto: Evan Ackerman/IEEE Spectrum

These are “Anchorballs” that Team NCTU’s robots can deploy as mesh network nodes, but they’re also used as active landmarks to help with SLAM localization. They’re weighted at the bottom so that, after being dropped, they stop rolling with a camera pointing at the ceiling, and the robot can then correlate the image from the Anchorball camera with its own internal map. 


After a bit of communication trouble, Team NCTU decided that they needed to boost their system a bit, so they deployed their own tether of sorts yesterday.

nctuPhoto: Evan Ackerman/IEEE Spectrum

If that looks like a router wrapped in a plastic bag connected to a really long Ethernet cord, well, you nailed it. The team placed the router on a robot, let the cable out from the base station as the robot drove into the mine, and then just gave the cable a tug to pull the router off the robot and onto the ground once the robot got far enough to deploy the router where the team wanted it.

Team CSIRO Data61

csiroPhoto: Navinda Kottege

These are some of the most cleverly designed droppable network nodes that we’ve seen—once dropped, the node waits for a few seconds for the robot to drive itself off, and then uses an actuator and springs to flip itself up and deploy its antennas. The reflective surfaces are a nice touch as well; presumably, they help DARPA find the dropped nodes at the end of a run, while also encouraging any following robots not to run over them quite as much.


irobotPhoto: Evan Ackerman/IEEE Spectrum

Why drop a network node that just sits there, when you could instead drop a network node that can drive itself around? Team CRETISE is dropping FirstLook robots from a beastly mothership robot to serve as cute little mobile nodes that can zip around to optimize your network.


The FirstLooks are themselves descendants of LANdroids, which were developed by iRobot with DARPA funding so long ago that I wrote about them back before I was even writing about robots, in early 2007. The original idea with LANdroids was that they would self-deploy to create dynamic and resilient mesh networks in challenging environments, but eventually iRobot turned them into the FirstLook tossable surveillance robots. And now, more than a decade later, Team CRETISE is turning them back into LANdroids again, which is pretty cool. 


marblePhoto: Evan Ackerman/IEEE Spectrum

In addition to little puck-shaped deployable mesh-network nodes, Team MARBLE has been experimenting with these robots built on top of remote-controlled off-road racing cars. They have onboard sensing and computing and communications, of course, but they’re not intended to search for artifacts. Instead, their job is to ferry information—quickly navigating between areas with base station connectivity and other (slower) robots that are exploring elsewhere.

Rather than acting as mobile nodes, these little cars are more like delivery robots, picking up data and carrying it back home. Without the constraints of having to maintain a network, the idea is that the exploration robots will be able to travel much farther while spending more time exploring autonomously, relying on the information-ferrying cars for all of their communication needs. The team doesn’t have it all working yet, but it’s a super interesting idea, and we’re looking forward to seeing it in action.

The Conversation (0)