Establishing security around airports, ports, or other infrastruc-
ture of economic or political importance is a challenge that is faced
today by police forces around the world. While randomized mon-
itoring (patrolling, checking, searching) is important — as adver-
saries can observe and exploit any predictability in launching an
attack — randomization must use different weighing functions to
reflect the complex costs and benefits of different police actions.
This paper describes a deployed agent assistant called ARMOR
that casts the monitoring problem as a Bayesian Stackelberg game,
where randomized schedule generation for police forces can ap-
propriately weigh the costs and benefits as well as uncertainty over
adversary types. ARMOR combines three key features: (i) it uses
the fastest known solver for Bayesian Stackelberg games called
DOBSS, where the dominant mixed strategies provide schedule
randomization; (ii) its mixed-initiative based interface allows users
to occasionally adjust or override the automated schedule based on
their local constraints; (iii) it alerts the users in casemixed-initiative
overrides appear to degrade the overall desired randomization. AR-
MOR has been successfully deployed at the Los Angeles Inter-