DARPA Announces Tasks for DRC Finals

Robot faceplants guaranteed!

5 min read
DARPA Announces Tasks for DRC Finals
Photo: DARPA

Late last week, DARPA released the final rules document for the DRC Finals. This document includes the list of tasks that robots will have to compete to score points, and we’ll take you through the details that are available of all eight of them.

For the DRC Trials in 2013, DARPA released an enormously long document describing every single minute detail of each task, from the precise orientation of the cinder blocks on the terrain task to the force required to turn each valve in the valve task.

We were expecting something similar for the DRC Finals, with the exception of the “surprise” task, but it’s looking like DARPA has decided to not provide anywhere close to the level of detail that allowed some teams (like SCHAFT) to complete most tasks on their own, ahead of time. Instead, all that we have so far is a brief description of each task, and here they are, in their entirety. Seriously, entirety.

  1. Drive the vehicle (same vehicle type as in Trials)
  2. Egress from the vehicle (get out of the vehicle)
  3. Open door and travel through opening. The door will open inward (away from the robot). The door will not include a threshold. Once fully opened, the door is designed to remain open.
  4. Open valve (similar to one of the three valves in Trials). DARPA will use a circular handle with a diameter between 4 inches (10 cm) and 16 inches (40 cm). The valve opens by counter-clockwise rotation. 
  5. Use a cutting tool to cut a hole in a wall (similar to one of the two tools and the wall in Trials). A circle will be drawn on the wall, approximately 8 inches (20 cm) in diameter. The cutting operation must entirely remove all wall material from the designated circle. 
  6. Surprise manipulation task (not disclosed until Finals).  The task will require manipulation and no mobility.
  7. Traverse rubble - Either cross debris field (by moving the debris or traversing it, similar to Trials) or negotiate irregular terrain (similar to Trials)
  8. Climb stairs (fewer steps and less steep than in Trials). The stairway has a rail
    on the left side and no rail on the right side.

So for example, with the first task, we have no idea how far the robots will have to drive the vehicle, whether they’ll have to avoid obstacles, or anything else. We know that the robots will have to drive a vehicle. That’s it. Same with the rest of the tasks, meaning that the Finals are going to be a total surprise to everybody, and teams won’t have a chance to practice any specifics.

We asked about this just to make sure that we’re not missing anything, and DARPA Public Affairs told us that yes, everything has been left intentionally vague, and nobody is getting any additional information. Furthermore, the courses will be changed slightly over the course of the Finals, to prevent teams from just using replay approaches to improve their score on the second run. The idea here is to try and give the teams as much of an experience of an unknown situation has is realistic for them to tackle using currently available technology.

Here’s a flowchart illustrating the order in which the robots will perform the tasks; note that the driving task in particular is optional, as is the robot getting out of the vehicle by itself (egress). Teams may skip these tasks and the points that come with them by walking (or otherwise locomoting) their robot around the driving portion of the course.

img

The tasks aren’t all mandatory (teams can skip some), but since scoring is primarily based on task completion (and only secondarily on time), there’s no reason not to try to make it through everything.

The rules document that DARPA released did give some additional details on how the competition will take place. Here are the highlights:

  • The maximum time allotted for a run is 60 minutes.
  • Teams will be allowed two runs, meaning two attempts to perform the tasks. Each run will include all of the tasks (unlike the DRC Trials, in which each task was attempted separately). Teams may attempt different tasks on different runs; that is, the runs are independent of each other.
  • For all DRC Finals tasks, unlike for the DRC Trials, the robot may not use a safety belay, may not use a power tether, and may not use wired communications with the Operator.
  • If a robot falls during a run, it may continue if it can resume the task without physical intervention, or if the team requests a Reset. Robots that have durability and the capability of getting up from a fall are encouraged.
  • If a robot is unable to progress, a team may request a Reset. Once a Reset is declared, at least 10 minutes must elapse before resuming the run. [The only penalty for a reset is this 10-minute penalty, but since the maximum time allotted for eight tasks is 60 minutes, your robot had better be pretty damn quick if it’s going to need more than one reset.]
  • Teams may make temporary modifications to the vehicle provided that they require no more than 5 minutes to install, require no tools, are passive (neither require nor provide power), and do not damage the vehicle. Approved examples include temporary handles taped on to the steering wheel, and a replacement seat that fits into the same connectors as the original seats. Disallowed examples include a second robot for driving, a generator or battery to provide additional power, and an actuator on the wheel or gas pedal. 
  • A robot may reconfigure its hardware during a run, but may not receive human help to do so. Teams may reconfigure a robot’s hardware between runs. 
  • Robots may use passive tools and leave them behind at any point during a run. For example, a robot may use Ski Poles.
  • Teams may not use a UAV or a throwable or launchable device. For safety, airborne devices are not permitted. [Aww.]
  • To promote autonomy DARPA will create controlled network service interruptions with substantial duration in which only low data-rate communication between humans and robots will be possible. The blackouts will be structured so that teams with more autonomous systems will be able to progress through a run more quickly. However, there will also be sufficient data exchange during some periods so that teams with less autonomy will be able to perform some tasks successfully, albeit more slowly.
  • At the start of a run no blackouts will last more than 30 seconds. The average length of blackouts will decrease during the course of a run. At about 45 minutes into a run, blackouts will no longer occur. The blackout schedule for a given day will be made available to teams the previous day. The same schedule will be used on all courses and for all runs through a given day. The schedule may change from day to day.
  • Cloud services, connected to the Operator Control Station by a Virtual Private Network (VPN) provided by DARPA, are optional and may be used for increased computing power and storage. The data rate to the cloud services will be 50 Mbit/sec in each direction. 

I especially like this question that one of the teams asked DARPA after reading through this rule set:

Teams:  We suggest that during the race, a belay could be attached to the robot, provided that the team was willing to accept a penalty on the scores or time.
DARPA:  Belays will not be allowed.

POW! Robot faceplants guaranteed.

The first look that anyone who doesn’t work at DARPA will get at the course will be on June 2, when teams will get a walkthrough, followed by a rehearsal day on June 4. The Finals themselves run June 5 and 6, and in case you’ve forgotten, they’re free and open to the public. We’ll see you there!

[ DRC Finals ]

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
LightGreen

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