​

A GUARD PATROL is nearly pointless if it’s predictable, because criminals and terrorists usually surveil potential targets as they plan their attacks. That’s why many security managers direct their officers to vary their routines. But what security professionals may not know is that humans trying to randomize actions invariably sink into predictable patterns, unintentionally offering adversaries the predictability they seek.

A new software program called Assistant for Randomized Monitoring over Routes (ARMOR) may solve the problem. Developed by faculty and students at the University of Southern California (USC), ARMOR has a decision-making algorithm that allows the program to create truly random routes for guard patrols while still achieving the desired coverage.

ARMOR grew out of Praveen Paruchuri’s 2007 doctoral thesis. He analyzed how to apply game theory to “multiagent” systems, in this case an environment in which agents, adversaries, and technology interact.

In his thesis, “Keep the Adversary Guessing: Agent Security by Policy Randomization,” Paruchuri took the example of a humanitarian relief mission, guarded by an unmanned aerial vehicle (UAV) like those flying over battlefields in Iraq and Afghanistan. The mission’s geography is divided into standard grid squares, and each one—including supply routes and storage warehouses—is assigned a risk value.

Guided by a randomization algorithm that incorporates risk data, a UAV could autonomously patrol the landscape. The algorithm could order it to monitor risky areas more frequently.

Paruchuri’s advisor at the Viterbi School of Engineering, Milind Tambe, discussed his student’s work on computerized randomization at a meeting of USC’s Center for Risk and Economic Analysis of Terrorism Events (CREATE). Erroll Southers, an assistant chief with the Los Angeles World Airports Authority (LAWA) police, which serves Los Angeles International Airport (LAX), is also a faculty member and is CREATE’s associate director. He thought all of it was extremely relevant to what they were doing at LAX. “They thought they could use it in the real world,” Tambe told Security Management. At the time, patrol elements, such as temporary vehicle checkpoints on airport access roads, were often moved predictably from week to week.

Tambe—along with some USC students and researchers from University of Maryland; Israel’s Bar-Ilan University; and Paruchuri, now with Intelligent Automation, Inc.— customized ARMOR’s software for LAX. The data points considered by the algorithm include site configuration, patrol locations, staffing levels, and threat information.

ARMOR’s user interface features a simple Microsoft Excel-style schedule, with shaded fields representing suggested sites for motor vehicle checkpoints and explosives detection canine patrols. Tambe emphasizes that ARMOR is intended to act as an assistant that recommends scheduling and locations. It is not intended to dictate patrol scheduling.

ARMOR at LAX was tested with a six-month pilot of the customized program, which ended earlier this year. That first iteration was not without wrinkles. LAX police and the developers found that during a given week during the test, one access road was only covered by a checkpoint 5 percent of the time, compared to 25 percent for another route.

Fixing the problem was simple, however, says James Butts, LAWA’s deputy executive director for public safety. It merely involved tweaking the software. The test was judged a success, and the software is now a part of LAX security.

Butts says the Transportation Security Administration (TSA) has expressed interest in ARMOR.

The program is currently also under consideration as a candidate for funding by the Pentagon’s Defense Advanced Research Projects Agency.