Military Robot Creators Look to Add More Autonomy, Soldier Collaboration
PITTSBURGH – For years, unmanned robots have been working with the military on the battlefield – think about unmanned drones or ground units that can disarm an IED – but so far they’ve been remotely operated and have been vulnerable to communications. Moving forward, researchers are looking to build more autonomy into the robots, as well as better collaboration between robots and soldiers.
Topics like this were discussed recently during a 10-year celebration of the Robotics Collaborative Technology Alliance, held in Pittsburgh at Carnegie Mellon University’s Robotics Lab. Discussions during the day focused on how the RCTA is developing fully autonomous robots that will partner with the warfighter in the field.
The goal is for robots to “possess a common understanding of the environment with the soldier and conduct independent maneuvers,” said Dr. Stuart Young with the U.S. Army Combat Capabilities Development Command (CCDC). The RCTA has three main objectives, Young said – robots that can maneuver in unstructured and hostile environments; robots that team with human soldiers; and robots that can manipulate their surroundings. Robots need to have strong situational awareness and be able to communicate with the soldier through hand signals and gestures.
Future military robots will need “intelligence, perception and robustness,” said Dr. Philip Perconti, director of the CCDC. Dr. Martial Herbert, dean of the School of Computer Science at CMU, and director of the National Robotics Engineering Center, agreed with that assessment. “They must be able to optimally utilize whatever processing power they have at any given time,” he said, and be able to “discern when it makes sense to approach a target via a less-than-optimal route to avoid enemy contact, the same way a soldier does.” Young said another goal is to develop robots that can “combat, process, and win against near-peer adversaries.”
Working with soldiers
Robots in the future will require less soldier oversight as possible, to give warfighters leverage. It currently takes two soldiers to operate a robot, said Kevin McEnery, deputy director of the Army’s Next Generation Combat Vehicle Cross Functional Team, which is modernizing its combat vehicles. The goal is to have a dozen robots operated by a single human being.
Researchers are adding artificial intelligence and machine learning capabilities to robots because “it’s impossible to hard code every single scenario they might face,” said Percanti. Another speaker elaborated on this difficulty – “If I drop this microphone, no one will be surprised when it falls to the ground,” said Dr. Dieter Fox, professor of computer science and engineering at the University of Washington. “Understanding this concept has actually proven very difficult for robots.”
Speakers discussed how much autonomy robots will need in order to be useful, as well as how the Army will get soldiers to trust robots, especially when it counts. Lt. Col. Christopher Lowrance, assistant professor and deputy director of the electrical engineering program at West Point, said that “because the Army can’t rely on networks in the field to push data, these robots need to be as autonomous as possible.”
Another speaker, Lt. Col. Jay Wisham, an RCTA program manager, agreed, but added that there’s a “minimum threshold that could be very beneficial to the warfighter and save lives. For a team of 10 to 15 soldiers on a platform, the data a robot provides can be a matter of life and death. Even if it isn’t connected to the network, and can’t download complete battlefield information, the information that the robot gives the team can still be enormously beneficial.”
Different tasks require different amounts of bandwidth, said Dr. Ethan Stump of the Army Research Laboratory – it’s important for researchers to “find the best and optimal middle ground.”
Even if soldiers don’t understand how robots make decisions, the Army is working on ways to get soldiers to trust the robots. “It often comes down to soldiers interacting with the system very early on, so they trust its actions, know its limitations, and feel when something is wrong,” said Lowrance. Fox agreed, saying “humans often aren’t particularly good at explaining how we come to our decisions, either.” Wisham added, “Simple things will earn trust first, while complex things will take more time.”
The unstructured battlefield
While companies have been strong adopters of robots over the last decade, the have almost all been deployed in highly controlled environments, such as a manufacturing floor or warehouse. “We can control everything from the temperature to the lighting,” said Dr. Siddhartha S. Srinivasa, professor of computer science and engineering at the University of Washington. By contrast, military robots will need to be robust enough to operate not only in highly unstructured environments, but also in places where enemies might be actively trying to destroy them.
“As a commander, I want the ability to decide when, where, and on what terms to engage the enemy,” said Wisham. “I want to be able to decide whether to make first contact with a human or a robot. I want to have that choice.”
New problems to solve
RCTA researchers at the event said they are excited about the collaboration with the Army. “The opportunity to work on the highest priority problems of national defense” gives researchers opportunities to get exposure to problems they wouldn’t see elsewhere, said Dr. Nicholas Roy, professor of aeronautics at MIT. “Knowing how to get rid of a lot of the assumptions that exist in a lot of commercial systems and put them into systems in the military context is hugely valuable for the students.”
Srinivasa agreed: “We’re able to educate an entire new generation of Ph.D. students, master’s students, undergrads, and young faculty, to think about what it is to solve problems out in the field. I think we’re growing the seeds of trees that will grow and really be the future of robotics for the United States.”