New DARPA Grand Challenge for Humanoid Robots -- Preliminary (Unofficial) Details

New DARPA Humanoid Grand Challenge

It seems we're going to have a new DARPA Grand Challenge!  The BAA with formal details should be out very soon, but for now we're bringing you the unofficial, preliminary details based on notes from Dr. Gill Pratt's talk at DTRA Industry Day: The new Grand Challenge is for a humanoid robot (with a bias toward bipedal designs) that can be used in rough terrain and for industrial disasters.  The robot will be required to maneuver into and drive an open-frame vehicle (eg. tractor), proceed to a building and dismount, ingress through a locked door using a key, traverse a 100 meter rubble-strewn hallway, climb a ladder, locate a leaking pipe and seal it by closing off a nearby valve, and then replace a faulty pump to resume normal operations -- all semi-autonomously with just "supervisory teleoperation."  That's a tough challenge, but it should be fun!  It looks like there will be six hardware teams to develop new robots, and twelve software teams using a common platform (PETMAN anyone?!).  The most crazy part about all of this: The United States is getting back into the humanoid robot game... in a big way!  Updated 4/10/2012 with official details!


Official Details (This Section Added 4/10/2012):


Today DARPA provided the official details in DARPA-BAA-12-39 Robotics Challenge:

The primary goal of the DARPA Robotics Challenge program is to develop ground robotic capabilities to execute complex tasks in dangerous, degraded, human-engineered environments. The program will focus on robots that can utilize available human tools, ranging from hand tools to vehicles. The program aims to advance the key robotic technologies of supervised autonomy, mounted mobility, dismounted mobility, dexterity, strength, and platform endurance. Supervised autonomy will be developed to allow robot control by non-expert operators, to lower operator workload, and to allow effective operation despite low fidelity (low bandwidth, high latency, intermittent) communications.  See the attached file "DARPA-BAA-12-39 Robotics Challenge" for the full text of the solicitation. The attached Broad Agency Announcement (BAA) is not a draft and represents the FINAL solicitation for the DARPA Robotics Challenge program.

DARPA Humanoid Grand Challenge


Here are the details for the task, as mentioned in the full-text details (PDF):

In the DARPA Robotics Challenge, robots will compete with each other performing disaster response operations in representative scenarios that will likely include the following sequence of events:

  1. Drive a utility vehicle at the site.
  2. Travel dismounted across rubble.
  3. Remove debris blocking an entryway.
  4. Open a door and enter a building.
  5. Climb an industrial ladder and traverse an industrial walkway.
  6. Use a tool to break through a concrete panel (see Figure 1).
  7. Locate and close a valve near a leaking pipe (see Figure 1).
  8. Replace a component such as a cooling pump.


Our original report was surprisingly accurate.  Oh, and Hizook scooped everybody -- from the New York Times, to IEEE Spectrum, to DARPA itself.  The moral of the story... keep reading Hizook -- your premier source for (informed) robotics news!  ;-)

PS -- Seriously, a special thanks to the folks who contacted us about the new DARPA program.  Your attention makes Hizook possible and meaningful.  Also, thank you to all the news sites that picked up the story from Hizook and provided attribution (no mention in the NYT feature, regrettably).  That latter aspect (attribution) has been a severe pain point for us as of late, as mainstream news sites (including blogs) have been providing piss-poor attribution while "jacking" our stories.  If you see it happening... be sure to give 'em hell for us.

Now, back to the original article.


Quick Background


For some time, I have heard musings about a new DARPA program in the wake of M3, ARM-S, and ARM-H.  These three programs are all led by Dr. Gill Pratt, and have resulted in numerous high-profile robot projects, such as: inflatable robots, the ARM-H robot, and the Robot Cheetah.  These are likely just the tip of the iceberg, as the programs are still underway.

Inflatable Robot from Otherlab  Inflatable robot arm from Otherlab  DARPA ARM-H humanoid  Cheetah robot from BDI

Now... Gill is a tried-and-true roboticist (he used to run the MIT Leg Lab).  So I'm sure the success of DARPA's previous Grand Challenges (autonomous cars) were not lost on him:  We need something like that for mobile manipulation.  And well... this is DARPA, after all, so they're going big​.



The New Grand Challenge


Through the grapevine, I heard that Gill announced an impending Grand Challenge at DTRA's Industry Day, so I reached out to a few folks to get details.  Kent Massey, from HDT Robotics, supplied us with the most comprehensive recount of Gill's announcement, that he provides with one major caveat:

Please understand that the content of this message is based on my own notes from the [DTRA Industry Day] meeting, and may not be totally accurate. The BAA will be out shortly and will have much more information.


Right.  So take all this with a grain of salt... but we should receive confirmation in a few days.  Kent says (emphasis mine):

At a DTRA Industry Day on Wednesday, Dr. Gill Pratt from DARPA announced a new Grand Challenge. Dr. Pratt warned that the BAA will be released within a week or two, so he will not be available to make further comments, other than what he disclosed at the DTRA Industry Day.

This new Grand Challenge is for a humanoid robot that can be used in rough terrain and industrial disasters. DAPRA will fund six hardware teams and twelve software teams. Although companies and organizations can bid on both hardware and software, they will, at most, only receive one award, either hardware or software, based on which of their proposals is stronger. There will also be opportunities for unpaid hardware and software competitors.

The BAA will heavily bias the robot morphology towards bipedal, although it's not clear that a humanoid form factor will be an absolute requirement. DARPA is contracting separately with a vendor to produce a humanoid robot that will be government furnished equipment (GFE) for the software teams. The vendor of the GFE will not be allowed to compete in the Grand Challenge. The GFE will likely be tethered to a power supply.

The software teams are expected to be able to function on the other hardware platforms, although it is no t clear to what level that will be a requirement. At some point, although this was also not clear, software teams may choose to focus on one or more particular hardware options.

DARPA will separately fund the development of a simulation environment and a model of the GFE. Dr. Pratt's hope is that this simulation environment will be sufficiently robust that it will become an industry standard. DARPA will supply 100 high-end workstations on a cloud for un-paid software teams to work on control of the GFE model in this environment. The unpaid software teams will compete against the 12 paid software teams. If some of the unpaid teams perform better than the paid teams, they will displace the paid teams and begin receiving funding.

DARPA will also invite unpaid hardware teams to participate. Dr. Pratt envisions that some foreign competitors may choose this option, due to restrict ions involved in accepting DoD funding. This Grand Challenge competition will have no ITAR restrictions and will be completely open to any participants. Unpaid hardware participants may also choose to provide subsystems to other paid and unpaid hardware competitors.

The specific tasks are:

1) The robot will maneuver to a open frame utility vehicle, such as a John Deere Gator or a Polaris Ranger. The robot is to get into the driver's seat and drive it to a specified location.

2) The robot is to get out of the vehicle, maneuver to a locked door, unlock it with a key, open the door, and go inside.

3) The robot will traverse a 100 meter, rubble strewn hallway.

4) At the end of the hallway, the robot will climb an ladder.

5) The robot will locate a pipe that is leaking a yellow-colored gas (non-toxic, non-corrosive). The robot will then identify a valve that will seal the pipe and actuate that valve, sealing the pipe.

6) The robot will locate a broken pump and replace it.


The robot will be teleoperated, at least at the supervisory level. DARPA will control the communications bandwidth and latency, in order to make the task more difficult and force higher levels of autonomous behavior. If necessary, this control over communications will be used to discriminate performance levels between competitors and select a winner.

The Grand Challenge will be run twice, one year apart. Given the extreme difficulty of the challenge, it is very likely that no robot will be successful in the first year.

DARPA will provide the paid teams with sufficient funds to do the work. Raising outside money or large contributions of IR&D should not be necessary.

The goal of this Grand Challenge is to create a humanoid robot that can operate in an environment built for people and use tools made for people. The specific challenge is built around an industrial disaster response.


-- Kent Massey

As a commercial plug for my company, HDT Robotics, I would recommend our arms and hands as an excellent starting point. We come much closer than anyone else at matching human dexterity, strength, speed, and form factor. We provide that level of performance in a rugged system at a very affordable price.



Your Thoughts?!  My Random Musings


Well, this is pretty awesome!  It's also a huge surprise.  The US has largely turned its back on legged humanoid robots over the last two decades (unlike Japan).  I actually thought this was a good​ thing, particularly for service and home robots, but perhaps the military perspective is altogether different?  This is sort of ironic given that Japanese roboticists are (somewhat) refocusing on non-legged robots in the wake of the Fukushima embarrassment.  [For those not in the know, Japanese roboticists have been chided by the government for their inability to apply robots in the disaster.  Furthermore, there was some embarrassment when iRobot, a foreign company, stepped in to lend robotic assistance. ] 

Certainly, some things have changed in the last two decades: computational capabilities, control algorithms, perception, batteries, and actuators.  But is the time really ripe to solve this problem now?  Frankly... if any knows, it would be Gill Pratt.  He was a driving force in the Leg Lab's foundational work.  He knows his stuff!

I'm looking forward to the new hardware systems.  Obviously, Boston Dynamics (BDI) will be a big player, especially for bipedal legs (like PETMAN, see below).  I think Meka Robotics, best known for their dexterous arms, could be a contender with their new legged humanoid -- which benefits from more commonplace electromechanical actuators rather than pneumatic or hydraulic actuators.   I know of a few other (research-level) groups that will probably be relevant... but who else could be one of the six hardware teams?  Finally, I hope DARPA funds at least one non-bipedal entry.  I'm a big proponent of "whatever works" rather than "human-like at all costs."  Frankly, I like my robots to be distinctly non-human... (yes, yes, I'm aware of the "we build our environments for humans, so leverage those affordances" argument).   Anyway, I'd love to see some quadrapeds like BigDogAlphaDog, or even an inflatable robot system.

Meka Robotics humanoid   AlphaDog Robot from Boston Dynamics  Big Dog Robot from Boston Dynamics


I think it's pretty clear... the common platform will almost certainly be some variant of PETMAN (perhaps with more dexterous arms).  There just isn't any other bipedal robot on the planet right now that could even come close for this task.

PETMAN humanoid robot from Boston Dynamics  PETMAN humanoid robot from Boston Dynamics  PETMAN humanoid robot from Boston Dynamics  PETMAN humanoid robot from Boston Dynamics


Will Boston Dynamics remove themselves from the hardware and software programs by becoming the GFE supplier of the common platform?  I doubt it.  Remember... it was RE2 who did all the integration work on the ARM-H common platform using Barrett arms.  Perhaps we'll see another integrator step in to combine components?


Anyway, we don't have much yet in the way of an official announcement... but I thought this was too exciting to not share.  Let me know your thoughts in the comments.  I'll be sure to udpate the post when the actual BAA is released.




PS -- Thanks to Troy Packer for the thumbnail image via his "colouring a robot in photoshop" tutorial.



I don't think there's a need to make manipulation more difficult with bipedal robots.  The last program, DARPA ARMS, didn't exactly meet its goals.  In the original Grand Challenge, they had to have the challenge twice before any teams could complete the course and that's for a "solved" and well studied area of robotics.  If they want a bunch of hacked together demos that don't generalize well, sure, of course you'll likely be able to make something but real progress is much harder.  I'll just leave the comment with this rant from the Berlin summit report:

"Robotics has been exposed! Witness the tasks in figure 3.2 that top researchers in the US had to do in a very structured environment (table top, good lighting, perfectly positioned stereo rig and ranging sensor) with a state-of-the-art manipulator and arm. Six teams competed. The best team completed task 1 four of five times with an average completion time of 136.1 seconds. The best team completed task 2 four out of five times with an average completion time of 153.7 seconds. The best teams completed task 3 two out of five times with average completion times of 84 and 274 seconds. Three of six teams either did not attempt or complete this task. After 50 years, shouldn’t these tasks be homework problems in a robotics undergraduate class? Arguably the simple environment in figure 3.2 is a structured environment and industrial robots have been used successfully in such environments. Why is this task so difficult?"


I have received additional reports that (indeed) my details may not be 100% accurate, but that the crux of this article is spot-on.  One contact, who wishes to remain anonymous, said this:

Just wanted to let you know one piece of information that was not reflected in your article. As the same, this is not official info. so take it with a gain of salt...

* Additional to the tasks you have described, the robot will have to pick up a concrete cutting tool and make a stitch cut in the wall and break through.

Making the task even more difficult...  ;-)

—Travis Deyle


"the robot is to get into the driver's seat and drive it to a specified location."

Now that there are intelligent cars that can pretty much drive themselves, why need a humanoid to drive a car?? An automous car was a previous -and successful- DARPA challenge anyway...



I was also at the DTRA Industry Briefing and, with the same 'grain of salt' disclaimer...

DARPA is striving to address some of the lessons learned from earlier challenges, including those mentioned by another commenter. As Kent mentioned, they have already planned and budgeted for two events, a year apart, and they are fully funding a significant number of teams, thus avoiding the distraction of fund raising (and commensurate 'hacking together of demos' often caused by insufficient funding).

The question of  'why not just use an autonomous vehicle in the scenario' was also asked at the session and Dr. Pratt pointed out (something along the lines of) that when disaster strikes, we want the robotic system to be able to use whatever resources are at hand, including vehicles, tools, instruments, spare parts, etc. Also, the vehicle is a minor aspect of the challenge scenario (ingress from the staging area to the facility) and the fun really begins once the robotic system leaves the vehicle and enters the damaged facility.

—Bob Touchton

I look forward to seeing the BAA.  I'm not going to be involved with this, though; one Grand Challenge entry (2005) was enough for me.

The state of the art is probably far enough along that throwing money at the problem will work.  But it will take a lot of cash. 

Legged locomotion is more or less understood for the not too hard cases.  If you have a humanoid robot with palm and knee pads, crawling is an option, which is probably a reasonable way to get through an obstacle-strewn hallway.

Getting into an open vehicle has never been done. But for something like a Polaris Ranger, the machine can grab onto the roll cage for the move into the vehicle.  Once in, driving can be more or less teleoperated, with the humanoid robot acting mostly as a servo system.

Opening a door with a key has been demonstrated by DARPA's "arm" manipulation project.

Climbing a ladder is mostly a mechanical problem. If both the hands and feet are articulated enough and strong enough to grab rungs, it should be do-able.

Grabbing and turning a valve shouldn't be too hard.

Replacing a pump - now that's tough. Assume the pump is a common industrial pump. There will be two threaded pipe or hose connections, several mounting bolts, and several electrical connections.  Robotic manipulation of threaded fasteners in unstructured environments has not been very successful.  People have been struggling with that for over 40 years now, and we're still seeing theses on the peg-in-hole problem. In controlled factory environments, with specialized tooling, fasteners are routinely inserted and tightened by robots.  Almost always, though, there's a clear approach to the fastener from the front, and almost always, the other end of the fastener is captive. Always, the fastener is brand new.  Without those advantages, the problem is much tougher.

The oil well ROV crowd routinely uses impact wrenches with their teleoperators. They definitely deal with plumbing. This isn't a mechanical problem.  Doing it autonomously, though, is new.

This is a problem that should first be solved by DARPA's "arm" project. When that two arms and a torso machine can replace a pump, the technology can be transferred to a mobile robot. I've seen a video of it picking up a screwdriver, so at least someone is thinking about this.  This is a problem which really should have been solved by now, considering all the money thrown at things like the NASA Flight Telerobotic Servicer. It's basic to being able to do automated maintenance of anything.

John Nagle

The value in having a single robot run through a large array of different tools and tasks is to show it is functional enough to replace a human for an unpredictable task.  A car is an interesting tool in that it drastically augments the robots own state rather than (mostly) the world state.

The ability for a robot to identify available tools (and even better identify things that could be adapted into tools), plan around the expected behavior/effects of the tool, and adapt it's own state control methods to properly control the tool are the near impossible parts here.  They are also the parts that are really interesting.  I think everyone believes it's possible to construct a robot which, mechanically, can perform the required movements.  DARPA has almost conceded this by providing the simulation/no hardware required options.

In my mind what DARPA is looking for is not new motors, actuators, sensors, or devices of any kind.  Instead they are looking for innovative control strategies.