Visualizing safer events

U. S. Department of Homeland Security uses agent-based simulations and video games to train and plan for public protection.

VBS2 (CT-Analyst) screen shot depicting people reacting to toxic cloud.

By Douglas A. Samuelson

As the U. S. election campaigns intensify, and Americans are bombarded with information and opinions about the election, we hear little about protection of the candidates and the crowds who come to see them. This is good, as this subject tends to make the news only when there is a calamitous failure. OR/MS and related techniques, especially visualization, play a key role in maintaining the security we have enjoyed.

The U. S. Secret Service (USSS), a component of the Department of Homeland Security, is responsible for, among other duties, protecting key dignitaries: the president, the vice president, major presidential candidates during the campaigns and their families. In the past decade, with increasing concern about possible attacks on large public events, USSS took on an expanding role in protecting not only the dignitaries, but also the crowds at some of these large events. By 2008, following up on experience from the national nominating conventions in 2004, USSS had acquired substantial modeling and simulation tools to help with planning and training for mass egress [Harmon, 2010].

For 2012, USSS’ parent agency, the Department of Homeland Security (DHS), undertook a new effort to expand and update the capabilities USSS had developed. A commissioned assessment of the state of the technology [Samuelson, 2011] indicated that the ideal goal – a single, integrated model and interfaces that could be used for large-scale planning, more local event simulations for training and real-time situation reporting – is not computationally feasible. Hence USSS and DHS elected to pursue, in parallel, two development tracks: a model of arena-sized movements, and a training and planning tool at the scale a single USSS agent would see during an event.

SSPT touch kiosk logo

Evacuation Planning Tool

The arena-scale planning for 2008 utilized the Evacuation Planning Tool (EPT), an agent-based simulation model of the venues in which the national nominating conventions were to be held [Harmon, 2010]. EPT was developed by Regal Decision Systems, which now markets a commercial version, SportEvac, that has been adopted by a number of professional sports franchises and stadium operators.

EPT/SportEvac is a noteworthy improvement in depiction over the pioneering Homeland Security Institute/Redfish Group model [Samuelson et. al., 2007] that demonstrated the feasibility of modeling large-scale crowd movements in real time. The people are represented as stylized figures rather than as colored dots, and three-dimensional depictions are included (albeit at considerable loss of speed of execution.) People can be tagged with different colors to show whether or not they got out of the venue in a specified time interval. (The animation display can, at the user’s choice, be delayed until the computation is complete.)

EPT/SportEvac has provisions for adding more variety in human behaviors. The assessment [Samuelson, 2011] done to guide the recent effort had identified this task as a priority, as clearly the tendencies of people to try to remain in groups, to go back for straggling members of the group, to become fatigued and to vary in how they comply with directions are all important factors in how a mass egress will occur. However, Regal made limited progress with this aspect of the model in the course of the DHS project last year. Future enhancements no doubt depend on team owners’ and stadium operators’ enthusiasm for buying SportEvac.

SSPT screen shot seen on touch screen kiosk in training class.

Another high-priority desired enhancement is the ability to incorporate more detail of specific venues. Assessing contingency plans depends critically on accurate and precise information about potential trouble spots. At national nominating conventions, for example, seemingly minor decisions about where to place television cameras, cables and control booths can have major consequences. EPT/SportEvac has improved capabilities to take in CAD plans of a venue, where available, and to accept more detailed input by other means, also where available. However, some questions still require considerable human effort and judgment to produce a sufficiently accurate and detailed representation.

Yet another important desired step is the ability to interface with other models, especially some very good and well-tested models of toxic plume spread – in particular, CT-Analyst [Young et. al., 2004]. Few if any integrated models of crowd movement and toxic plumes exist. A critical issue in such interfaces is standardization or reconciliation of scales, units of measurement and nomenclature. Even more basic is the tendency of modelers to use digital input and pictorial output. Somehow the model passing data needs to convey digital information along with the visual, or the receiving model needs to be able to convert the pictures back into digits. Various approaches are current topics of DHS-sponsored research.

Validation against real events, which are admittedly and fortunately rare but do exist, was another desired step that remains for future research. Here, too, questions of scale, units and nomenclature pose substantial complications.

Regal is also investigating the limits of the EPT/SportEvac inference engine and considering what upgrade path to pursue. Ongoing technological improvements offer opportunities for better performance-depiction tradeoffs but may necessitate large development costs. Starting completely over may be more cost-effective than starting over with a new inference engine within existing data and depiction architectures. This is a promising area for future research that may not occur for quite a while under current budget conditions.

EPT depiction of evacuation of the Xcel Center

Site Security Planning Tool

Much of what USSS considered most urgent and important concerns the training and preparation of individual USSS agents, rather than plans for the venue as a whole. For example, effective crowd direction relies critically on the placement of some key people directing movement, and on their effectiveness and persuasiveness in providing such direction. Also, the individual is key to identifying major changes in the event: for example, detection of a colorless, odorless toxic plume, as in a nerve gas attack, would depend on someone recognizing the unexpected behavior (such as coughing and then falling down) of other people in his field of view.

The Security and Incident Modeling Lab (SIMLAB), a component of the primary training venue for USSS, has relied for many years on hands-on “around the table” exercises in “Tiny Town,” an HO-scale reproduction of a number of typical event venues, and on live human “game” exercises. The widespread availability of video games, and the large proportion of younger staff now familiar with such games, suggested a better approach. When USSS learned that the Royal Canadian Defense Force (RCDF) had, in fact, already developed a training simulation, CAMX (Civilian Activity Modeling for eXercises) [Levesque et. al., 2009], that could work within a video game engine such as JCATS (Joint Conflict And Tactical Simulation) or VBS2 (Virtual Battle Space 2), pursuing this technology for the current purpose looked very attractive. Accordingly, with the appropriate international and inter-agency agreements, USSS and its contractors worked with RCDF to develop the Site Security Planning Tool (SSPT).

Penn State University’s Applied Research Lab served as the technical lead in SSPT, creating the user touch interface with the host simulation/game. They also researched, tested and evaluated various hardware and software candidates that were incorporated into SSPT.

SSPT runs on large (55-inch) touch-screen kiosks mounted slanting toward the user in a cabinet. It presents either a third-person top-down (isometric) perspective for overview planning or a first-person “what you see from ground level” perspective for virtual walk-throughs of sites. These visualizations combine to give trainees a very realistic “look, sound and feel” experience in a highly realistic depiction of the venue of interest. Presenting and altering a scenario such as the nerve gas attack is fairly easy in SSPT. This enables USSS to replace other training methods, increasing effectiveness while reducing cost.

One accomplishment of the recent research was to integrate the pre-computed physics-based plume cloud transport/dispersion model output of CT-Analyst into SSPT via DIS interface to VBS2. In addition, corresponding health effects by the D-HHS/AHRQ are ingested into the plume cloud data, called a Nomagraf, which is imported by VBS2 and visually displayed in 3D according to wind direction and speed. CAMX helps provide the visual cues of the health effects, relative to exposure time and concentration, by dynamic behaviors representing symptoms. The USSS is now able to incorporate civilian clutter and resultant response measures (DCON, shelter-in-place, evacuation, etc.) in its training exercises pertinent to homeland security preparedness, incident response and consequence mitigation.

VBS2 (CT-Analyst) screen shot depicting toxic cloud in urban setting.

In addition, using SSPT, once a plan has been developed (whether it is a security, event or emergency plan), it may then be exercised in multiple games with rapidly updated features. Challenging the plan with certain threats reveals vulnerabilities, which then serve as teaching points to students or participants in the exercise. Analysis of the plan’s vulnerabilities leads to adjustments to the plan for additional execution and testing, decision-making and subsequent course of action analysis, coalescing the lessons learned from the training exercise together as a best practice.

As with EPT, SSPT presents a number of opportunities for improvements. The incorporation of more human behaviors is a high priority, along with the perennial concern for adding realism without unacceptably impairing speed of execution. In fact, for this kind of training, real-time responsiveness is even more required, and departures from it less tolerated, than in the larger-scale tool. Another desired feature is the development of real-time situation displays that closely resemble the SSPT displays used in training. (This appears to be an easier course of action than changing the SSPT displays to resemble some real-time situational display that may be developed.)

DHS, in ongoing collaboration and consultation with RCDF about the software, will continue to assess the effectiveness of training, learn what approaches are most effective and modify SSPT accordingly. As with EPT, the long-term goals are to dock SSPT with other models, incorporate additional effects, facilitate detailed input of additional venues and more readily explore new training ideas.

EPT depictions of Evacuation of Fort Meade High School (Maryland)

Next Steps

The past decade has seen a proliferation of modeling efforts in pedestrian dynamics, mass egress, and related areas [for example, see Kuligoswki et. al., 2005 and 2010.]. Just listing them all, much less characterizing strengths and weaknesses, is a daunting challenge. Integrating these modeling efforts, working toward standardization and compatibility, and recognizing what works and what doesn’t, both in the modeling and in the depicted practices and policies, remains a rich area for future research – that is, no one is close to figuring it all out. Some priorities emerge, however, from what is now known.

More accurately depicting the variety of human behaviors and what influences them in these situations remains a critical need and a challenge. RCDF is conducting ongoing research to extract human behavior and its effects from a variety of real situations. Their 2011 publication summarizing work to date in this area is marked for limited release; the results promise to be of great interest to researchers and incident managers when they are made generally available. Inquiries to the authors of the earlier RCDF report [Levesque et. al., 2009] might prove fruitful for the respectably and respectfully persistent.

The effect of different kinds of signage is a subject in its relative infancy, with interesting results already [for instance, Veeraswamy et. al. and Xie et. al., 2009]. Medical effects are, in general, insufficiently understood and quite important. For example, medical experts disagree strongly about whether it is more effective to stabilize victims of smoke inhalation and transport them as soon as possible to a hospital to begin treatment, or to treat them on the spot with anti-inflammatory steroids [Loellgen, 2010]. This decision, in turn, affects the placement and movement of medical treatment personnel and victims at the scene, which in turn affects preferred egress routes.

Interactions at different scales, and with different modeling focus, remain another challenge. There are excellent models of the spread of toxic plumes (e.g. Young et. al., 2004], but they’ve had little success so far in docking plume models with movement models. Good models of egress from a stadium rarely dock with models of pedestrian and vehicle movement in the surrounding community.

Studies and plans regarding movement, in turn, too rarely interact with analyses of site design and resource placement. Stadium egress models indicate that inflow of emergency responders and equipment substantially affects egress. This fact and its implications are routinely taken into account in some building designs; it is quite common, for example, for hotels and large office buildings to have built-in standpipes, hoses, and, in many cases, pumps and sprinkler systems, greatly improving the speed and effectiveness of response to a fire in comparison to having to bring in hose lines. Sports arenas, however, tend to have less emergency equipment for the number of people at risk. More system-level studies are called-for, even before the modeling software reaches the capability to simulate these questions.

A recent Homeland Security Institute study, based on extensive input from first responders [Royal et. al., 2012], identified readily accessible, high-fidelity virtual training tools as a top priority need for first responders at this time. Also deemed critical were real-time identification of new risks, real-time tracking of where responders are and whether they are at risk from developing hazards, real-time monitoring of availability and functionality of resources, and the ability to identify trends, patterns and important content from multiple sources, including non-traditional sources, to support incident decision-making. In short, there is an increasingly widespread perception of growing need for the kinds of tools and methods USSS and now DHS have been developing.

VBS2 (as used in CMAX) depiction of toxic cloud from CT-Analyst


Recent efforts by the U. S. Secret Service and its parent agency, the Department of Homeland Security, have produced interesting improvements in modeling overall planning of mass egress for large venues and for high-fidelity real-time training tools for incident managers and responders. The Evacuation Planning Tool (EPT) is now being commercially marketed, as SportEvac, to professional sports teams and venue operators. Many such private-sector and local public-sector operators are now interested in better incident planning, and opportunities in this area appear likely to remain promising. EPT/SportEvac may be nearing some limits in its ability to increase fidelity while maintaining fast response, however, so additional research and development opportunities appear likely. The Site Security Planning Tool (SSPT), based on a video game and implemented via touch-screen kiosks, appears to represent a very promising approach that others will want to emulate or implement directly.

USSS and DHS used these tools in protecting the national nominating conventions and will continue to apply them for the presidential inauguration and other large events. Simulation and visualization tools have demonstrated excellent results in providing better site security planning, “what-if” assessments and training at reduced cost.

Douglas A. Samuelson ( is president of InfoLogix, Inc., a small R&D and consulting company in Annandale, Va., and a senior statistical subject matter expert for Great-Circle Technologies in Chantilly, Va. Great-Circle does pattern analyses for the intelligence community and other federal clients. For Homeland Security Institute, in 2006-2007, he led the team, including Redfish Group, that produced one of the pioneering agent-based real-time simulations of mass egress from a large venue, PNC Park in Pittsburgh, Pa. Samuelson was a senior expert consultant to the 2010-2011 U.S. Secret Service effort described here, taking a key role in assessing what work should be done and how the pieces of the project might fit together. He was co-general chair of the PED 2010 conference and is internationally known for his work on this subject. He is a contributing editor of OR/MS Today. He has a doctorate in operations research from The George Washington University.


  1. M. Harmon, M., J. Joseph, 2010, “Evacuation Planning Tool (EPT) for Emergency Event and Space Planning,” in Peacock, et. al., eds., “Pedestrian Dynamics and Evacuation” (PED2010), Springer, New York, N.Y.
  2. E. Kuligowski, R. Peacock, 2005, “Review of building evacuation models,” NIST TN 1471, National Institute of Standards and Technology, Gaithersburg, Md.
  3. E. Kuligowski, R. Peacock, B. Hoskins, 2010, “Review of building evacuation models,” NIST TN 1471, 2nd Ed., National Institute of Standards and Technology, Gaithersburg, Md.
  4. J. Levesque, F. Cazzolato, J. Martonosi, 2009, “CAMX: Civilian Activity Modelling for eXercises and eXperimentation: Development and trial of a software prototype,” Defence R&D Canada, Technical Memorandum, DRDC CORA TM 2009-XXX, Kingston, Canada.
  5. H. Loellgen, 2010, “Injury of Lung and Heart After Inhalation of Toxic Substances,” in Klingsch, et. al., “Pedestrian Dynamics and Evacuation 2008” (PED2008), Springer, New York, N.Y.
  6. M. Royal, E. Goldstein, et. al., 2012, “Project Responder 3: Toward the First Responder of the Future,” U. S. Department of Homeland Security, HSSAI Document Number RP10-68-94, Homeland Security Studies and Analysis Institute, Arlington, Va.
  7. Douglas A. Samuelson, M. Parker, L. Miller, A. Zimmerman, H. R. Blacksten, S. Pommerenck, 2007, “Agent-Based Simulation of Mass Egress After an Improvised Explosive Device Attack,” Homeland Security Institute Final Report to the Science and Technology Directorate, U. S. Department of Homeland Security, HSI Document Number RP06-IOA-31-03, Homeland Security Institute, Arlington, Va.
  8. Douglas A. Samuelson, M. Parker, A. Zimmerman, S. Guerin, J. Thorp, O. Densmore, 2007, “Agent-Based Simulations of Mass Egress After an Improvised Explosive Device Attack,” Proceedings of the European Conference on Complex Systems, October 2007, Dresden, Germany.
  9. Douglas A. Samuelson, 2011, “Agent-Based Simulations of Mass Egress: Summary of Current Capabilities and Recommended Next Steps,” Technical Report to the U. S. Secret Service, Lionheart Publishing, Atlanta, Ga.
  10. Douglas A. Samuelson, 2012, “Agent-Based Simulations of Pedestrian Movement for Site Security: U. S. Secret Service’s Current Capabilities and Next Steps,” in Schreckenberg, et. al., eds., “Pedestrian Dynamics and Evacuation”  (PED2012), Springer, New York, N.Y. (to appear).
  11. A. Veeraswamy, E. R. Galea, and P. Lawrence, 2009, “Implementation of Cognitive Mapping, Spatial Representation and Wayfinding Behaviours of People within Evacuation Modelling Tools,” Proceedings of the 4th International Symposium on Human Behaviour in Fire, Robinson College, Cambridge, U.K., pp. 501-512, July 2009.
  12. H. Xie, L. Filippidis, E. Galea, D. Blackshields, P. J. Lawrence, 2009, “Experimental Study of the Effectiveness of Emergency Signage,” Proceedings of the 4th International Symposium on Human Behaviour in Fire, Robinson College, Cambridge, U.K, pp. 289-300, July 2009.
  13. 13. T. Young, Jr., J. Boris, S. Hooper, C. Lind, K. Obenschain, G. Patnaik, 2004, “Emergency Assessment with Sensors and Buildings: Advances in CT-Analyst Technology,” resented at CBIS 2004, Williamsburg Va., October 2004. Laboratory for Computational Physics and Fluid Dynamics U.S. Naval Research Laboratory, Code 6400, Washington, D.C.