Rapid Transit Systems such as metro or bus lines constitute prime targets for terrorism and criminal activity. The key reason for this is the large number of passengers that systems such as the LA Metro or the Chicago transit system carry every day (e.g. approximately 350000 people for the LA Metro and 750000 for the Chicago transit system).
Hence, the intelligent deployment of human resources such as officers and canine units to patrol such systems constitutes a key challenge for several security agencies. In particular, avoiding predictable schedules is extremely important. Any predictable pattern could be exploited by criminals or terrorists to attack a bus or metro line. To avoid this, we developed TRUSTS a decision-aid tool to produce accurate randomized patrol schedules for each patrol officer. TRUSTS is composed of two main components: (i) a main system which computes patrol schedules and a mobile app that each officer can use to request a patrol schedule or to submit information about the checks that he made during his shift. In what follows, we discuss each component in detail.
TRUSTS: Scheduling Randomized Patrols for Fare Evasion Deterrence in Transit System
TRUSTS has been designed to produce effective patrol schedules to deter fare evasion within the Los Angeles Metro system (see the Figure). TRUSTS models the patrolling problem as a Bayesian Stackelberg game where patrollers act as defenders and fare evaders as attackers of the system. Thus far, our research has investigated ways to produce accurate, scalable and robust approaches that could adapt and handle the uncertainty and the unpredictability of a transit system. The results of this research were published in a number of papers at several international conferences. We refer the interested reader to the list down below to download the papers. Currently, we are investigating ways to extend our system to incorporate crime prevention and counter-terrorism.
LA Metro Map
The Mobile Application
TWe have developed a mobile app that any patrol officer can use to request a schedule from TRUSTS. The app consists of a user interface composed of three different views that each officer can refer to while patrolling (see Figures below). Each view offers a different functionality: (i) the schedule view can be used to request and update a patrol schedule; (ii) the report view can be used to submit information about each check that has been made and (iii) the summary view can be used to verify the information collected within each shift. Currently, we are testing our mobile application in collaboration with the Los Angeles Sheriff Department.
Security Games in the Field
One of the key research achievements of the TRUSTS project is that we were able to conduct controlled experiments of our game theoretic resource allocation algorithms. Before this project, the actual evaluation of the deployed security games applications in the field was a major open challenge. The reasons were twofold. First, previous applications focused on counter-terrorism, therefore controlled experiments against real adversaries in the field were not feasible. Second, the number of practical constraints related to real-world deployments limited the ability of researchers to conduct head-to-head comparisons
In TRUSTS we were able to address this challenge and run the largest scale evaluation of security games in the field in terms of duration and number of security officials deployed. We evaluated each component of the system (Fare Evasion, Counter Terrorism and Crime algorithms) by designing and running field experiments. In the context of fare evasion, we ran a 21-day experiment, where we compared schedules generated using game theory against competing schedules comprised of a random scheduler augmented with officers providing real-time knowledge of the current situation. Our results showed that our schedules led to statistically significant improvements over the competing schedules, despite the fact that the latter were improved with real-time knowledge. For counter-terrorism, we organized a full-scale exercise (FSE), in which 80 security officers (divided into 14 teams) patrolled 10 stations of a metro line for 12 hours. The purpose of the exercise was a head-to-head comparison of the game-theoretic scheduler against humans. We compared the schedule generation process and had the performance of both schedules evaluated by a number of security experts. Our results showed that game-theoretic schedules performed at least equivalently to (and in fact better than those) generated by human schedulers. Finally, we ran a two-day proof-of-concept experiment on crime where two teams of officers patrolled 14 stations of a train line for two hours. Our results validate our OSG model in the real world, thus showing its potential to combat crime.
Co-authors and Collaborators