Super human: The process of building a brighter robot

Even for the less fanciful among us it is difficult to greet the news that the world's biggest defence contractor is working on a six-foot, 300lb humanoid robot with autonomous capabilities without at least a flicker of alarm.

After all, memories of the 'Terminator' films are now so culturally ingrained that they are hard to shake. And, as we know, the advent of such machines in those films did not end well for the human race as a whole, so we can perhaps be forgiven the odd shiver. However, all is not as sinister as it might seem. In fact, the Atlas Robot was not developed by Lockheed Martin, being a robotic hardware platform developed by Boston Dynamics with US Defense Advanced Research Projects Agency (DARPA). It is a hydraulically-powered robot in the form of an adult human capable of a variety of natural movements, including dynamic walking, calisthenics and user-programmed behaviour. It has 28 hydraulically-actuated joints with closed-loop position & force control Atlas was modified to meet the needs of the DARPA Robotics Challenge. The DARPA Robotic Challenge is predicated on a disaster response scenario and focuses on developing robots that can operate in rough terrain and difficult conditions, using aids (vehicles and hand tools) commonly available in populated areas. More specifically, the Challenge is designed to prove that the following capabilities can be accomplished: • Compatibility with environments engineered for humans (even if they are degraded) • Ability to use a diverse assortment of tools engineered for humans (from screwdrivers to vehicles) • Ability to be supervised by humans who have had little to no robotics training. A number of teams entered this competition, going on to compete in a 'Virtual Challenge' in which their control software was applied to certain tasks in a simulation environment. These tasks included: walking over increasingly difficult terrain; picking up a hose from a table and connecting it to a standpipe; and finally, approaching an all-terrain vehicle, use the robotic platform to drive through a course and then exiting the vehicle. One of the teams to have passed this stage and to have taken delivery of the Atlas robot was from Lockheed Martin's Advanced Technology Labs in conjunction with the University of Pennsylvania and Rensselaer Polytechnic Institute. And, as Bill Borgia, director of the Intelligent Robotics Laboratory at Lockheed Martin Advanced Technology Laboratories points out, this chimed with much of the defence company's existing research in this area. "One of our long-term interests is to develop robots that can operate in highly-complex, human-engineered environments…The Challenge aligned very well with what we were working on already," he says. "We see robotic systems expanding into environments that are much more complex and much less predictable than those in which they currently operate. The kinds of technologies required to compete in the Robotic Challenge are very, very well aligned with where our organisation is going." Naturally, having to adapt its technology to fit a platform such as Atlas that Lockheed did not develop is a challenge, but one that Borgia sees as a positive. "We don't know what the robots that will be developed in the future are going to look like. This is simply the first of many platforms where we will apply our techniques to controlling. As we expose our technology to additional platforms, we'll learn more and more about how to control them." 'Supervised autonomy' is a critical element of the Challenge, as it allows simple tasks to be performed by the robot without full-time operator intervention. This will be especially important in unreliable communications environments. Says Borgia: "The level of autonomy is heavily dependent on the nature of the task. There are some tasks that have a high level of autonomy because they're very predictable, but there are other tasks where the human has to collaborate with the robot to help it understand what outcome the human desires. "So, for example, tasks like walking over flat terrain the robot can achieve very well by itself. However, others, such as understanding its environment and what it looks like, are less autonomous at this point and require collaboration." The tasks outlined by DARPA's challenge involve robots going into a complex industrial facility and using human tools to perform actions that humans could otherwise not perform. So, for example, they are looking at tasks such as clearing rubble, opening doors, climbing stairs and ladders and breaking through a concrete barrier. Clearly these tasks present a lot of opportunities for developing new technologies. For example, building a robot that can pick up and operate a drill involves a lot of close co-operation between sensing, perception, planning and control of the robot. However, according to Borgia, the key to the development process does not lie in simply attempting to anticipate the tasks the robot will be asked to perform and programming it accordingly. He says: "If we try to anticipate all the scenarios, we'll ultimately end up building a solution that works in those scenarios, but may prove not to be adaptable to others. We're in the business of building systems that we want to be broadly applicable." One of the biggest issues facing the team is therefore to identify what the right trade-offs are between anticipating specific scenarios and trying to build a system that is highly adaptable. Says Borgia: "One of the things we do as engineers is to identify instances that may exist across multiple scenarios and implementing them in certain ways whereby they are broadly applicable. "So, for example, turning cylindrical objects such as valves comes up a number of times in the robotic challenge. But the process of opening valves is not that different to turning a steering wheel in the driving task. Those processes are broadly similar and we can aggregate them in such a way that the same technology can be applied." Of course, ever-greater autonomy for the robot is a key aim in the longer term and, by familiarising the robot with certain tasks and allowing it to adapt these capabilities in alternative scenarios, ever greater autonomy becomes feasible. "The more linkages we can make between potential tasks, the higher the capability that exists for performing those tasks autonomously," says Borgia. Nonetheless, the heart of the project remains the collaboration between the robot and the human operator. "Ultimately," says Borgia, "the human and the robot will work together not to replace the human, but to enhance the human's capability and becoming a trusted remote partner to accomplish tasks he could otherwise not perform. "It's a highly dynamic and unpredictable environment, so however well we might try to model that environment, there are still going to be objects and situations that we didn't anticipate. Those situations will also change over time, so the robot can plan some of those actions upfront, but the human remains a crucial partner to help the robot understand the outcomes that will complete the mission." In terms of the likely level of autonomy he believes the robot may be able to achieve, Borgia insists that will be intrinsic to how critical such autonomy is to the mission in hand. He says: "Ultimately, our goal is to help then operator to accomplish his mission. If there are cases where the operator cannot do that by himself, we want to be able to apply autonomy to accomplish his mission." Of course, the big question remaining is over the areas in which such technology may be applied. Given Lockheed Martin and DARPA's involvement, it doesn't take a great leap of imagination to suggest that this technology's ultimate destination will be military in nature, but this is by no means the only application. Says Borgia: "As a technologist, I'm fascinated by the potential multitude of applications. I liken it to the PC revolution. It was difficult for the engineers of that time to foresee all the potential applications, but once they got them into the hands of the users, the application space really exploded. "With the Challenge, we'll put the Atlas into the hands of trained disaster responders, but as it continues to become exposed, many more domain experts will be able to identify potential applications for the technology going forward."