How do you address a hostile threat or secure a perimeter in a constantly-changing environment?
Perhaps an enemy boat is about to cross a protected border, or a tanker needs to dodge obstacles in a docking port.
Current solutions are scripted and decision-making is centralized. The ideal, most resource-effective approach, says Gaurav Sukhatme, a professor in the Viterbi School's Department of Computer Science, would deploy teams of humans and robots for just the right mix of sensing and problem solving.
Viterbi CS professor and lead investigator Gaurav Sukhatme with a robot prototype.
The Office of Naval Research has awarded Sukhatme and his collaborators at USC and three other universities a $7.5 million Multidisciplinary University Research Initiative (MURI) grant to develop a foundation for such an ideal response.
“MURI challenges state a very clear focus and a clear national need,” says Sukhatme. “The objective is also clear: we’re aiming to design algorithms and solutions that revolutionize the way things are done.”
Entitled ANTIDOTE: Adaptive Networks for Threat and Intrusion Detection Or Termination, the five-year MURI effort will aim to develop a basis for networked coordination and control that can be applied to any number of contexts — including autonomous vehicles, sensor networks and teams of humans and robots.
The ANTIDOTE program is the first to tackle fundamental distributed control across such a variety of networked entities. Existing efforts have addressed these issues in isolation.
Sukhatme is the MURI’s lead principal investigator, and USC is the lead university. Viterbi professors Maja Matarić and Sven Koenig (both also of CS) will collaborate.
The USC team will work with researchers from Carnegie Mellon University, the University of Pennsylvania and the Massachusetts Institute of Technology.
“We’re excited to be the leading university in what will be groundbreaking research,” says Viterbi Dean Yannis Yortsos. “MURI awards provide an excellent opportunity for groundbreaking research because they support large, multiple university collaborations over a five-year grant period, and as such, enable real progress on hard research problems.”
The Viterbi team brings internationally-renowned expertise to the task. Sukhatme is an expert on multi-robot coordination with a specialty in underwater robotics and robot networks. Matarić is one of the world’s authorities on robot teams and on human-robot interaction, while Koenig specializes in robot navigation and planning for both single robots and robot teams.
“It will take a tight collaboration of the whole MURI team to develop integrated solutions to the whole problem,” says Matarić. “Our project considers the problem of robot and human teams at many levels, including navigation and planning and team coordination and human-robot interaction.”
Viterbi CS Professor Maja Matarić with one of her robots.
The scientists will study how best to deploy teams of humans and robots — working in conjunction with unattended sensors that collect data and take measurements in the field — to tackle a variety of functions. The approach must be effective, and the team must stay within communications range.
The key deliverable will be developing the algorithms that lay out the chain of command for humans and robots in a variety of scenarios.
“The robots must be able to react in real time,” says Koenig. “And because these scenarios can be highly dynamic, we need to develop fast-planning algorithms for them, which presents an additional challenge because planning with incomplete information is typically very difficult and thus time-consuming.”
For example, should one robot in the group discover a land mine, what sequence of instructions should that vehicle follow next?
“Does the robot peel off from the team, send a picture back to a human decision-maker, and hand off control?” says Sukhatme. “Or can we solve the problem without having a human at the joystick?”
In developing the algorithmic response, a number of problems must be addressed. For example, communication between team members may be unreliable; the behavior of the adversary might be unknown or unpredictable; and the environment may be dynamic. Members of the team itself might also have different capabilities; for example, one robot might be slower than another, yet have a better camera for sensing.
Viterbi CS Professor Sven Koenig.
“How do you design a system that makes the best use of these resources?” asks Sukhatme.
The MURI program will focus heavily on experimental work. Robot bodies and a laboratory on campus will be a testbed for the team’s algorithmic ideas. The program will employ up to five graduate students and three postdoctoral fellows.
Potential applications are vast; many involve force protection approaches for use by the military. Non-military applications include harbor safety and environmental monitoring.