Meet ATLAS!

12/03/2013 - 14:19

Robot to the rescue

Gill A. Pratt

Gill A. Pratt

Gill A. Pratt is a DARPA program manager on loan from Franklin W. Olin College, where he is a professor of electrical...

More

On March 12, 2011—the day after a powerful earthquake and tsunami struck the Fukushima Daiichi Nuclear Power Station—a team of plant workers set out to enter the darkened reactor buildings and manually vent accumulated hydrogen to the atmosphere. At first, the workers made progress inside the buildings, but soon their dosimeters showed they had reached their maximum emergency radiation exposure limits, and they had to turn back. In the days that followed, with vents still closed, hydrogen built up in each of three reactor buildings, fueling explosions that extensively damaged the facility, contaminated the environment, and drastically complicated stabilization and remediation of the site.

News of the earthquake galvanized those in charge of robotics programs at the US Defense Advanced Research Projects Agency (DARPA). Humanitarian assistance and disaster relief is a primary mission of the Defense Department, and DARPA had responded to a disaster before, sending robots whose development it had funded to New York City in the days after the 9/11 attacks. The robots found no survivors then, but perhaps this time, robots could help mitigate the evolving disaster.

Donate to the Bulletin and support reader-sponsored media.
be a donor

DARPA officials contacted researchers who had designed robots for the Three Mile Island and Chernobyl disasters and coordinated with companies that DARPA had funded to develop other robots—the iRobot PackBot, the QinetiQ TALON, and the Honeywell T-Hawk. The PackBot and TALON are man-portable ground robots originally developed for tactical reconnaissance. The T-Hawk is a tactical reconnaissance unmanned aerial vehicle. Each company was already making plans to send its robots and training personnel to Japan.

Although these robots were designed for austere environments, they were not designed for a radiological disaster. DARPA took on the task of projecting whether they would work or succumb to the radiation. The results looked promising: The robots’ materials and electronics would operate for at least several hours in all but the closest encounters with reactor fuel. It also seemed likely that the robot cameras would act as primitive “too close” warnings, displaying increasing amounts of “snow” before radiation levels got high enough to crash the robots’ computers.

The robots sent from the United States were joined in Japan by one from Tohoku University and eventually by several others from around the world. But weeks passed before power plant personnel completed training to operate the robots. By then, the best the robots could do was help survey the extensive damage that had already occurred and take radiation readings; the golden hours for early intervention to mitigate the extent of the disaster had long since passed. Officials at DARPA, and the robotics community in general, felt that a significant opportunity had been missed. In Japan, citizens and robotics experts reacted similarly, asking why companies that had developed wonderful industrial and entertainment robots had nothing to offer when disaster struck.

The world’s population is continuing to grow and move to cities situated along flood-prone coasts. The population over age 65 in the United States is forecast to increase from 13 percent to 20 percent by 2030, and the elderly require more help in emergency situations. Climate change and the growing threat of proliferation of weapons of mass destruction to non-state actors add to the concern. Today’s natural, man-made and mixed disasters might be only modest warnings of how vulnerable society is becoming. Robots failed to find survivors after 9/11 and ended up being of only long-term help at Fukushima. Could they do better?

Better technology through competition. Between 2004 and 2007, DARPA used a competition series—two Grand Challenges, followed by an Urban Challenge—to accelerate the development of technology for self-driving ground vehicles that would help to reduce the vulnerability of military logistics convoys to improvised explosive devices. In the wake of the Fukushima catastrophe, DARPA decided to take the same competitive approach with the DARPA Robotics Challenge (DRC), which aims to speed development of robotics technology for disaster response.

The overarching goal of the challenge is to develop technology that allows human beings to effectively intervene during a disaster, without being held back by the dangers such a situation poses to humans. In particular, the competition aims to improve the mobility and dexterity of robots working in degraded environments; to give robots the ability to use human tools, from screwdrivers to vehicles; and to drastically reduce the training needed to operate robots during emergencies.

After consulting with first responders and other disaster experts, DARPA officials selected a series of challenge tasks that require capabilities that might have been useful during the first days of the Fukushima response. The competing robots would have to drive a utility vehicle, exit, and walk a short distance. They would have to travel dismounted over rough terrain. They would need to clear rubble blocking a doorway, open a door, and go through it. They would have to climb a steep ladder. They would be able to cut an access hole in a wall with a power tool, without damaging nearby infrastructure. They would have to close an industrial valve. And they would need to pull a fire hose over a selected distance and connect it to a standpipe.

Following the successful model of its earlier challenges, DARPA broke the robotics competition into three sequential events. The first, the Virtual Robotics Challenge (VRC), was held in June 2013. A cloud-based, real-time simulation, the VRC had contestants from all over the world program and operate a standard, computer-simulated robot in the driving, rough terrain, and fire hose tasks just described. More than 100 teams and individuals initially signed up, with members of the top 10 performing teams coalescing to form seven software teams that would compete in the second event, the DRC Trials. Those seven software teams have either purchased or received through DARPA a physical humanoid robot called ATLAS, developed by Boston Dynamics with DARPA funding. All seven software teams were also provided with research funding to continue their software development work until the trials. (The software teams are led by the Florida-based Institute for Human and Machine Cognition, Worcester Polytechnic Institute, MIT, the Texas-based robotics and automation firm TRACLabs, Inc., Virginia-based TORC Robotics LLC, Lockheed Martin, and Hong Kong University.)

DARPA also funded seven systems teams through a proposal evaluation process to develop their own hardware and software systems to compete in the DRC Trials. One team was eliminated after design reviews, leaving six systems teams to compete in the DRC Trials, each with a physical robot of its own design. (These are led by Carnegie Mellon University’s National Robotics Engineering Center, Drexel University, NASA Jet Propulsion Lab, NASA Johnson Space Center, SCHAFT Robotics of Tokyo, and Virginia Tech.) In addition to the 13 funded teams mentioned above, DARPA is accepting teams without funding that wish to challenge the other robots. Based on the results of a qualification event in October 2013, a total of 17 teams will compete.

The DRC Trials were scheduled to occur at the Homestead-Miami Speedway in December 2013,  just weeks after publication of this article. Although attendance was limited, the event was open to the public and available through video updates provided via the DRC website.

Successful task completion is generally more important than speed in disaster scenarios, and DARPA expects the robots at the DRC Trials to operate very slowly, taking up to 30 minutes to perform each task. After two days of testing, up to eight of the best-performing teams will each be eligible for $1 million of DARPA research funding for one year, after which the agency will hold its third event, the DRC Finals, which will provide more difficult, faster-paced tests designed to mimic an actual disaster scenario. For example, the DRC Trials will be conducted with tethers to deliver electricity to the robots and safety lines to protect them from falling or otherwise damaging themselves; the DRC Finals will be conducted with neither. The winner will receive a $2 million prize.

Don’t be scared of the robot. Most of the public’s response to the first stage of the DARPA Robotics Challenge has been positive. Disasters make plain the fragility of human life, so systems that enhance individual protection and societal resilience are widely appreciated. Still, new technologies often raise broad moral, ethical, and societal questions, and DARPA encourages discussion among experts and stakeholders about the laws and policies that should apply to new capabilities developed through the agency’s efforts. Five such issues are particularly relevant in robotics, and DARPA carefully considered each one when designing the DRC.

First, although the DRC will not develop lethal or fully autonomous systems, some of the technology being developed in the competition may eventually be used in such systems. This is equally true for technology now being developed in civilian robots, for example, those meant to assist workers in factories or to aid the elderly. DARPA has worked diligently, particularly with those teams that are sensitive about accepting Defense Department funding, to honestly characterize the dual-use nature of the technology being developed in the DRC.

Second, society is now wrestling with moral and ethical issues raised by remotely operated unmanned aerial vehicles that enable reconnaissance and projection of lethal force from a great distance. While it is very difficult to predict the future use of any technology, the remote supervision technology being developed for the DARPA challenge is more immediately applicable to surgical robots than to such remotely operated aircraft.

Third, the tempo of modern warfare is escalating, generating a need for systems that can respond faster than human reflexes. The Defense Department has considered the most responsible way to develop autonomous technology, issuing a directive in November 2012 that carefully regulates the way robotic autonomy is developed and used in weapons systems. Even though DRC robots may look like those in the movies that are both lethal and autonomous, in fact they are neither.

The fourth issue involves the general impact of automation, which has been debated for decades. Like the Industrial Revolution’s impact on standards of living and working conditions, the development of robots will have both positive and negative effects. By analogy, the DARPA-developed ARPANET (which later became the Internet) has enabled significant gains in economic productivity, jobs, quality of life, and political freedom around the globe, but it has also made it easier for terrorists to learn how to build improvised explosive devices. DARPA believes the development of robots for disaster response represents a net gain for society because of the great danger to human life posed by disasters.

The final issue raised by the DRC has an emotional element. Disaster-response robots must operate in human environments, so they tend to be designed with limbs instead of wheels and tracks. This can give the robots a disturbing, not quite human appearance. Since the Golem of Prague, Western literature has included frightening tales of human-like monsters, with Frankenstein’s monster and the Terminator standing out as more recent examples. Being social creatures, human beings are skilled at observing and predicting one another’s behavior, so the first sight of a human-like robot with hard-to-discern intentions may induce fear. Disaster-response robots such as those involved with the DRC are specifically designed to avoid harm to humans, and history has shown that humans quickly learn to trust machines with proven track records of safety, like automobiles and airplanes.

While the technology in the DARPA Robotics Challenge will provide action-at-a-distance, with supervised autonomy of simple tasks, none of the robots being developed will act autonomously. Each robot will be supervised by one or more human operators, located out of sight of the robot, but connected via electronic communication. Because electronic communication is usually difficult in disasters, due to degraded infrastructure and higher-than-normal traffic (at Fukushima it was also made worse by heavy steel reinforcement of concrete and radiation shielding), the robotics challenge will purposefully vary the quality of communications infrastructure during events. Under these conditions, the challenge for the teams will be to provide the robots with just enough autonomy to execute small tasks, such as “take a step up the ladder” or “turn the valve clockwise,” without requiring continuous operator communications.

Still, DARPA expects the robots at the trials to demonstrate capabilities that extend beyond what robots could do at Fukushima. The DRC robots will be able to affect their environment, not just take sensor readings, and they will have just enough autonomy to deal with significantly degraded communications. But there likely will be many failures and very slow operation—robots with roughly the mobility and dexterity of a one-year-old child who can barely walk, falls down frequently, and can’t execute complex manipulation tasks. The trials provide 30 minutes for each task to be accomplished, and most of the comprehension of the scene will be done inside the brains of human operators looking at compressed robot sensor feeds, not independently by the robots.

Reducing the cost of robotic disaster relief. When DARPA ran its first Grand Challenge in 2004 for driverless vehicles, none made it past 7.3 miles of the 150-mile route. But a number of vehicles did succeed in the second Grand Challenge and the final Urban Challenge. Similarly, by the time of the DRC Finals, DARPA hopes the competing robots will demonstrate the mobility and dexterity competence of a 2-year-old child, in particular the ability to execute autonomous, short tasks such as “clear out the debris in front of you” or “close the valve,” regardless of outdoor lighting conditions and other variations. The robots will still need to be told by human operators which tasks to chain together to achieve larger goals, but DARPA hopes they will clearly demonstrate the impact robotics could have during the critical early hours of future disasters.

Vehicle manufacturers are now capitalizing on the earlier DARPA challenges by building cars that assist drivers or drive themselves. Similarly, the government and the private sector will need to engage in additional development efforts to maximize the disaster-mitigating potential demonstrated in the DARPA Robotics Challenge. Recently, Japan’s Ministry of Economy, Trade and Industry signed a collaboration agreement with the US Department of Defense in the field of disaster-response robotics, and the agencies are planning several collaborative workshops in the coming months, in hopes of inspiring other countries to join in developing technology to improve the world’s response to natural and man-made disasters.

Beyond governmental collaboration, private companies will have to step up to transfer the technology developed by the DRC into robots across a wide range of applications, from health care to agriculture to defense. Cost will be a prime consideration. Even though they are not hardened against truly dangerous environments, the robots competing in the challenge cost up to several million dollars each. To prepare for the next disaster, government and private industry must not only design disaster-response robots; they must reduce costs significantly so robots can be purchased and available before disaster strikes.

Because of the significant efforts of the Defense Department and manufacturers in the nine years since the first Grand Challenge, unmanned defense logistics convoys may soon preserve warfighter lives, and civilian driverless cars could also begin reducing the approximately 1.3 million automobile accident fatalities that occur each year worldwide. That is not to mention the effect self-directed autos would have on traffic jams and the quality of life of those (including the elderly) who cannot drive.

If manufacturers can harvest the technology that will emerge from the DARPA Robotics Challenge, these disaster-response systems could drive innovation in the health care, industrial, agricultural, and defense sectors, significantly improving the world’s common security and quality of life.

Editor's note: Research for this article was funded by the Defense Advanced Research Projects Agency.