WASHINGTON, July 11, 2013 – Leaps forward in simulation technology and cloud computing are making it possible for challengers from around the world to compete for support from the Defense Advanced Research Projects Agency to create robots that can help people during natural and other kinds of disasters.
During a recent media roundtable, Dr. Gill Pratt, DARPA program manager, and Dr. Brian Gerkey, chief executive officer for the Open Source Robotics Foundation, told reporters about the ongoing DARPA Robotics Challenge, which launched in October 2012 and will end after the final event in December 2014.
The Open Source Robotics Foundation is an independent nonprofit organization in Mountain View, Calif., founded by members of the global robotics community to support open-source software development and distribution for robotics research, education and product development.
“DARPA is focused on the defense mission for DOD,” Pratt said. “Our primary reason for [creating the robotics challenge] is about the security of our citizens in situations of natural and manmade disasters. [But] the technology DARPA develops often finds its way into all sorts of other parts of life.”
The Internet is the best example, he observed, adding, “I expect that the robots we develop will be used very soon, at least in some form, … within people’s homes,” possibly as helpers for aging populations in nations like the United States and Japan.
The goal of DARPA’s Robotics Challenge, or DRC, is to generate groundbreaking research and development in hardware and software, according to the DRC website, helping future robots perform the most hazardous jobs in disaster-response operations, along with human supervisors, to reduce casualties and save lives.
Pratt calls this a way to make societies worldwide more resilient to natural and other disasters.
“We believe it’s important to develop robots that can go into areas that are too dangerous for people and that can be supervised by human beings despite the fact that communications might be quite difficult [during a disaster] both between human beings and between people and robots,” he added.
Through the DRC project, DARPA is helping create robots with three basic features. The first is that the robots should be compatible with environments engineered for people, Pratt said.
“That’s true even if those environments have been degraded. This gives the robots a certain size [and] it says exactly what their capabilities must be in order to interface for instance with doors and stairs and other things that human beings have engineered into the environment,” the DARPA program manager said.
The second feature is that the robots have to be able to use tools that were designed for people.
“This ranges all the way from a screwdriver to a fire truck,” he added, “so you’ll see in the different parts of the DARPA challenge that we are testing the ability of these machines to do that.”
Third is that the robots must be able to be supervised by people who aren’t necessarily trained to operate robots.
“Typically in a disaster there’s no time for training [and] there’s no time to acquire specialized tools. You have to use what you have on hand,” Pratt explained.
“If you want the robot to respond immediately,” he added, “the important thing is for the interface between that person and the robot to be intuitive to the people on the disaster response team who have the most expertise about what needs to be done … not the people who designed the machine.”
Overriding all the robot technology, though, is an assumption that communications between people — and between people and the robot — will be degraded by the effects of the disaster on infrastructure, Pratt said.
“So in the challenge itself,” he explained, “we will purposefully lower the bandwidth — the number of bits per second — that can go between the robot and the supervisors, and we will also increase the latency — the amount of time delay — in the communication between the people and the robots.”
The DRC has two kinds of events — one for teams whose focus is software alone and that don’t have their own robots, and one for teams whose focus is both hardware and software and therefore have their own robots. The first event, whose seven winners were announced June 27, was a software competition among 26 teams from eight countries.
The teams competed against each other using a virtual robot called Atlas inside the DARPA Robotics Challenge Simulator, an open-source tool created for DARPA by the Open Source Robotics Foundation.
A company called Boston Dynamics is using DARPA funding to build real Atlas robots that the winning teams have been awarded to use in upcoming DARPA challenges.
“Our reason for having this virtual challenge is that we wanted to open the contest to teams that were strong not just in building hardware for robots and programming them with software,” Pratt said, “but for a wider variety of teams, including those who had little expertise or experience with robot hardware.”
Pratt said work done by the Open Source Robotics Foundation with DARPA funding had advanced the simulation technology enough that the simulator could run in real time and a person could interact with the simulation to supervise the virtual robots.
The Foundation’s approach to simulation is to do the best possible job of reproducing the way physics works in the world inside a computer, Gerkey said.
Thanks to the increase in performance that allows the simulation to run in real time and the increased computational power available through advances in cloud computing resources, the seven teams who won the DARPA virtual challenge should be able to take the software they designed for the DARPA simulation and run it on the real Atlas robots.
“Our goal,” Gerkey said, “is always to have the simulator behave as close as possible to the physical system … so it should be the case that teams who … are awarded an Atlas robot from Boston Dynamics should be able to take the software that they develop for simulation and run it almost unchanged on the physical robot.”
He added, “That will actually be the test of how well we’ve done here in terms of building a simulation as a stand-in for the physical robot.”
All of what DARPA and the Open Source Robotics Foundation are building is open source, Gerkey noted, “so anyone in the world can do anything they want with this software.”
The next DRC live competition will be held in December 2013.