DARPA’s Virtual Robotics Challenge a practical step forward on the long road toward disaster-response capable humanoids
The Fukushima Daiichi Nuclear disaster was a wake-up call to the robotics community. In Japan, many asked why a country known for its cutting-edge robotics sector was unable to respond to the emergency. Worldwide, robotics experts pointed to the event as a real-world test of what robots can and cannot do.
Whether man-made or natural — or like Fukushima, a combination of the two — major catastrophic events, while rare, are becoming increasingly costly as human populations worldwide move to urban areas. This is why, in an effort to spur the development of agile humanoid first-responders, the US Department of Defense’s strategic plan identifies disaster response as a priority area, and why it is funnelling tens of millions of dollars into the DARPA Robotics Challenge.
DRC Program Manager Dr. Gill Pratt envisions a future where tele-operated robots can be quickly deployed by disaster-response personnel with minimal robotics training: “In Fukushima, during the first 24 hours, the explosions would not have occurred if it would have been possible to vent the reactors,” said Pratt in a media round-table last month. “Human beings tried to do this but had to turn around [because of the radiation]. Our hope is to develop machines that can actually intervene and help make a disaster [like this one] less severe.” Estimates of the total economic loss due to the Fukushima disaster range from $250B – $500B US.
Why a humanoid?
Though having a robot that can safely navigate dangerous places like the inside of a nuclear plant is no-doubt useful, one has to question whether the extreme cost and design challenge of developing a humanoid that is capable of doing so is worth the effort. Yet as Pratt points out, a disaster scenario that is catastrophic to humans is also most likely to occur in an environment that has been engineered for humans, and what kind of robot would be more capable of navigating in a human environment and using the tools found there than a humanoid?
As DARPA project leaders were designing parameters for the Robotics Challenge, they consulted with first-response experts — including the Nuclear Regulatory Commission — and identified key capability requirements that indeed seem to point to a humanoid solution: the ability to go through doors, climb steep stairs, walk on rough terrain, and the ability to use whatever tools are oh hand, such as a utility vehicle or a screw driver.
“Our belief is that we will never know where the next disaster is going to hit and when it will be, and so we have to be prepared to use whatever tools are on hand,” said Pratt. “And at least for the foreseeable future, those are tools for people to use, and so we need to adapt our machines to use those tools as quickly as possible.”
Designing a roadmap for success
DARPA does not define the success of its programs through a profit and loss model, and its goal with the DARPA Robotics Challenge is not to bring a humanoid first-responder to every fire station in the country. As a government agency, DARPA focuses on bringing high-risk technologies to the point of prototype, and then handing these off to the military or commercial sector for development. It invests in challenges such as the DRC in order to spur radical innovation, and short-term technical failure is OK as long as the long-term potential payoff is high.
But despite these lofty goals, the question remains: what’s the roadmap for designing and overseeing a major robotics challenge such as this one? How does one gauge success when the potential payoff may be decades or more away?
DARPA’s previous autonomous car challenges offer some lessons, and one of them is that there isn’t always a winner. Pratt admits that success for the DRC isn’t guaranteed, but says that the goal of the competition isn’t so much to declare a winner as it is to continually nudge the competitors towards greater achievements, while at the same time being realistic about what is actually doable. As a result, DARPA will be reevaluating the project’s milestones and adjusting the difficulty of the tasks in consultation with the participants along the way.
Previous autonomous car competitions are different from the current challenge in one major aspect, however: the older competitions were based on automotive technology, which already had a significant infrastructure in place. The robotics industry, by comparison, is still in its infancy, and so a great many hurdles exist for competitors, both at the software level and at the hardware level, and in terms of integration. Pratt has thus broken the competition down into three stages and four separate competitor tracks in an attempt to structure what is otherwise an extremely open-ended problem. (You can find a good explanation of how the competition breaks down on DARPA Robotics Challenge Fact Sheet.)
The first stage of the DRC — the Virtual Robotics Challenge (VRC) — offers particular insight into DARPA’s strategy for success. The VRC, for which the results were announced earlier this month, was a cloud-based competition designed to test software, and its goal was to evaluate the teams’ robot perception, manipulation and locomotion in a virtual environment as a first step before moving on to the physical trials, which will take place in December of 2013. The big idea here is that teams will be able to try out their strategies in simulation first, before investing in hardware. “The reason we had this virtual challenge is that we wanted to open the contest to teams whose strength is primarily software,” said Pratt.
DARPA’s simulator models real-world physics and behaviours with incredibly high accuracy. Brian Gerkey, CEO of the Open Source Robotics Foundation (OSFR), the organization behind the VRC simulator, says that their goal is for the simulation to be the best possible stand-in for the real robot, so that teams that have done well in the VRC should be able to run their software virtually unchanged on the physical robot during the trials. And yet run to such a complex simulation — and to play that simulation back at interactive rates — takes a huge amount of computer power.
“One of the key developments over the past several years has been the advancement of cloud computing resources,” explained Gerkey. “It’s now possible to rent access to cloud computing resources so that you can have modest computer at your end and have the simulation run somewhere else. That’s a key aspect of this project and it’s really required in order to run the competition.”
The seven winning teams from the VRC challenge each received an Atlas furnished by Boston Dynamics for use in the DRC trials, where they will compete with other funded and unfunded teams that have been developing their own hardware (see DARPA’s Meet the Teams video to get a sense of what Atlas will be up against). They now have until late December to transfer their software algorithms to the Atlas hardware shell, and practice running a real machine in real environment. The physical trials should be a good demonstrator of the success of the simulator approach to competition design.
Building a legacy
DARPA takes a long view to technology development, starting with identifying a major national security need (in this case: disaster response), then determining the specific technologies and innovations needed to address that need (in this case: easily deployed agile humanoid robots), and finally working to support the fundamental breakthroughs required to make these technologies possible. DARPA’s disaster challenge is thus about creating infrastructure to support long-term software and hardware development in the robotics community; it’s also about attracting bright minds to solve some of the world’s most pressing problems.
One of the legacies of the VRC is that all the software code for the simulator is being distributed open source, and so, as Gerkey points out, there is no cost to reuse it later.
“The history of software development has shown that this is the best possible, most efficient development model for building the common platform that a whole community comes to rely upon,” says Gerkey. “The simulation work that we’re doing is really part of a broader effort to build a common ecosystem of software tools and libraries that everyone can use. There are computers that are powerful enough to run this simulation that you can rent for on the order of a couple of dollars an hour.”
The hope is that, ultimately, the open-source VRC simulator will help speed up the development of robots in general by lowering the barrier of entry and that this, in turn, will lead to more capable robots – perhaps ones capable enough to be of service in a Fukushima-like disaster. Says Gerkey: “Once we can run this very realistic simulation in a competition, we can then try this in a classroom scenario. You don’t have to outfit your classroom with expensive powerful computers.”
Pratt agrees: “Right now we have a few really talented designers and a few companies that are beginning to be strong in the robotics field. We would like to see those numbers go way up, and we believe that the way to do it is to give people a tool to develop these machines without having to incur the cost of first building them.”
Fostering international cooperation
Fukushima is used as an example to motivate the need for robotic platforms in dangerous environments, and the VRC is being promoted as an internationally beneficial endeavour, so I asked Pratt what specifically is being done to foster international cooperation, and whether there was a possibility that DARPA would work with other countries on future challenges.
Pratt answered that a key measure for success in this project is engagement with the international community. “All the code that we’re writing is distributed under the Apache 2 licence which is a very permissive open source license that allows anyone in the world to do anything that they want with this software. We would love for the challenge (once we are done with it in the next year and a half) to move around the world, and we have been in discussions with our counterparts in Japan and the EU about the possibility of this becoming a rotating contest. We think that [making societies more resilient to disasters] is a need that the whole world has, and so and I’m spending quite a lot of my time having lots of discussions about what the future of this is going to be once we are done … I personally think that the international part of this is a key way to measure how well we have done, so if we’re successful in having the DRC continue on somewhere else in the world, then it will be clear that we have done well.”
Though Pratt hesitates to predict how long it will be before we have robots capable of helping out in situations such as the one in Fukushima, he is convinced that DARPA’s Robotic Challenge will result in some major developments for the field:
Look out for DRC trial coverage in December!