ISI Leads R&D Effort to Build Computer Advisors for Combat Commanders

| March 8, 2005

The method uses committees of virtual advisors, animated by artificial intelligence, who share information and negotiate with each other to suggest solutions..

University of Southern California researchers have begun work on a $12 million computational control system to help combat commanders make quick, accurate decisions in critical situations.

The method uses committees of virtual advisors, animated by artificial intelligence, who share information and negotiate with each other to suggest solutions.Robert Neches and Pedro Szekely of the Information Sciences Institute, which is part of the USC VIterbi School of Engineering, are directors of the recently approved DARPA-funded project on “Criticality-Sensitive Coordination,” or CSC.

The ISI researchers will be working with engineers at Vanderbilt University’s Institute for Software Integrated Systems and the Palo Alto, California-based Kestrel Institute.

“The goal of CSC is to help soldiers and multiple tactical teams re-synchronize their mission plans when forced to adapt to changed situations. Criticality-Sensitive Coordination will enable them to respond more rapidly and more accurately,” said Neches, director of ISI’s Distributed Scaleable Systems division.

“Generations of officers have known how difficult it is to issue the right orders in the right sequence under time pressure to accomplish tactical objectives, and the difficulty of coordinating changes in plans in a timely fashion to avoid developing problems,” Neches explained

“This effort is aimed directly at making sure that everyone stays in synch when you shift to alternate plans, so no one ‘zigs’ when the other guy is ‘zagging’ – something that sounds easy but is tremendously hard to do when time is short and bullets are flying,” he continued.

The CSC approach to coordination is what Neches describes as “dynamic partial centralization.”


Clustering for decision making: Affected unit’s assistant (red) makes contact with counterparts at other units (orange) who in turn reach out if needed, as needed, to more distant ones. The red unit assistant becomes committee chairman in this this process. click image to enlarge.

It begins by creating a detailed computer model of the forces in a command area; their tasks, schedules, capabilities, resources, dependencies on other units, status, and dozens of other relevant variables.

In the system, each unit or subunit is represented by a subprogram, a computer virtual “assistant.” These assistants can interact with each other.

“When units encounter a problem, such as delay, or inability to perform a task, the command system creates a ‘cluster,'” Szekely explained.

The assistants in the cluster form themselves into a committee. The most critically affected unit’s assistant becomes committee chair. This assistant retrieves all task and contingency plan information from the cluster members and computes a proposed solution.

The system then computes adjustments to the plans of all cluster members using centralized solver techniques and distribute the solution to all cluster members.


Resolving conflict: Two groups of units (green and blue) have formed clusters to deal with an emergency, leading to conflicting demands as the clusters overlap. The leaders negotiate with each other to combine. click image to enlarge.

If multiple clusters form, and have overlapping problems, the process is repeated, with each affected cluster negotiating, through its leader, with other clusters. (see diagram

This process can continue many steps. “Even then, it is still far faster than humans can evaluate consequences of possible actions,” Szekely said. “And it will still be humans who will act on the proposals presented.”

The system depends on effective and speedy machine- accessible communications that can track all the relevant variables in real time, Neches emphasizes. “But to a large degree, this information is now available. The problem is being able to comprehend the big picture quickly enough to make the right decisions. And we think that this system will be able to help.”

The CSC program is on a 4-year time track, during which researchers will start by using a relatively limited set of inputs, adding more and more as the system gains robustness.

The new system builds on a previous USC-Vanderbilt collaboration that created a consolidated flight operations and logistics system for military aircraft that is now receiving experimental use in Marine aviation units, and is expanding to a wider and wider circle of aircraft types under the terms of a $5.7 million contract awarded in 2003.

This earlier system creates and coordinates draft schedules for air operations of a squadron and for the supporting aircraft maintenance units, taking into account a huge range of needs and variables, including availability of pilots, mechanics, planes, runway slots, operation needs, weather and more.

“This earlier experience helped us understand the key principles and provided a number of critical lessons- learned which are inspiring the new work,” said Szekely.

Before the system, the task of drawing up such a schedule took many hours of skilled officer time. The well-tested system does the job – effectively – in a matter of minutes, and the Marine Corps is making plans to further develop it and deploy it throughout the Marine Corps.


Clustermakers: Pedro Szekely (left) and Robert Neches. click image to enlarge.

Published on March 8th, 2005

Last updated on August 9th, 2021

Share This Story