On the 6th August, a small protest in London quickly became an angry mob that overwhelmed the authorities. The public disorder spread like a wildfire into a full-scale country-wide riot.
Unfortunately, riots such as the ones in London can often escalate, leading to loss of human life. Now, researchers in Europe are using parallel computing resources to model unpredictable public order scenarios so police and military forces can be better prepared and to minimize future injuries for both civilians and law enforcement in urban environments.
“The main purpose of the project is to support a training process of security forces, such as military, policemen, etc. in asymmetric scenarios, where the forces in opposition are not conventional – one force has a structured hierarchy and rules of engagement, while the other is an unstructured group, using non conventional means and tactics.
Most often, there are less security forces than civilians; thus, they need special training to effectively handle this kind of scenario,” said Dariusz Król, a researcher from the Academic Computer Centre CYFRONET, which is part of the University of Science and Technology in Poland. CYFRONET is working in partnership with a dozen computing institutes in France, Germany, Slovakia, Slovenia and Sweden.
The work uses an amalgamation of parallel computing resources and the latest software simulations to understand the social complexities of crowds in tense situations as part of the European Urban Simulation for Asymmetric Scenarios (EUSAS). Asymmetric threats in urban territory involve an operation of a relatively small group of soldiers within a city with a civilian population ranging from neutral to hostile.
“Due to the large amount of possible civilians it is very difficult to predict the possible outcomes of different rules of engagement,” Król and his colleagues reported at the Cracow 11 Grid Workshop in Poland last week. “This task can be aided through a computer agent based simulation system, [however], in order to get meaningful data from a system, a significant number of runs with different parameters must be performed, where the combination of number of parameters and their ranges often exceeds 1010 possible combinations.”
The researchers reported a new way to manage these simulations by minimizing the number of simulation runs necessary and by monitoring simulations as they go, and even allowing researchers to change parameters on the fly.
The project may help to train real world personnel, such as the recently developed European Gendarmerie Force, a multi-European force made up of soldiers from five different countries that has militarized police functions and specializes in crisis management.
“More soldiers are involved in security missions, and thus must deal with crowd control in urban environments. But, their training doesn't include such unusual situations. The goal of the project is to provide a means for training in those environments, and the results are used to improve Tactics, Techniques and Procedures, that is, the way of managing such situations,” said Marc Contat, the EUSAS project coordinator from CASSIDIAN, a defense and security company.
To statistically understand if strategies for crowd management work, the researchers in Poland use a term called ‘Measure of Effectiveness’, which comprises a number of factors, including number of injured people and level of civilian aggression.
“Our statistical simulations include elements which are difficult to predict in the real world, for example, the number of civilians. Such elements can be included in the set of input parameters. Rules of engagement [for law enforcement] were implemented using official documents about this topic. As rules of engagement are well defined, we did not include them in the set of input parameters,” said Król.
The simulations include ‘virtual agents’ who are imbued with human characteristics, such as a particular level of aggression. The virtual agent toolkit used by the researchers is called MASON, an open source software that is designed to statistically model and visualize the complexities of social interaction. Król and his colleagues programmed their civilian and police agents with different logic variables or levels of emotion.
Scenarios were then run for a particular length of time with specified levels of inputs, such as how fearful are certain agents. One scenario can have 10 different input parameters, each of which has 10 different values. A one minute long scenario can be 5MB in size and there are 10 billion possible scenario combinations. This adds up to a staggering 50 petabytes of data.
The researchers use a distributed infrastructure that includes computing clusters, clouds, grids and high-performance computers to process all their data. Each computing platforms is combined to create a data farming environment. Hundreds of thousands of one-minute-long simulations are run simultaneously and each infrastructure is connected through ‘virtualization’, a method that enables an independent computing environment, complete with its operating system, to be run simultaneously within another.
Król said, “Unfortunately, even using clouds, grids and clusters together, we cannot explore all situations we would like to. Therefore, we provide a dedicated web interface to make simulations more interactive.”
Currently, the researchers are refining their models to make them faster and to include as many human behavioral factors and actions as possible. “However, none of these is possible if you have too many factors involved. Thus, we have introduced a number of techniques to make the complexity less painful,” said Król.
“We allow an experimenter to analyze partial results from finished simulations. Based on this information, they can decide what values of which input parameters should be explored more deeply, because they seem more promising than others. Also, we allow them to use different experimental methods to constrain the set of input values.”
This means that when a researcher wants to see how a specific set of parameters will pan out, they can run that simulation 20 times over. Then, after these 20 simulations are run, they can interactively alter the initial input parameters to see if a crowd’s aggression increases when a particular event occurs. According to Król this is an organic way of working and a first in his research field. Any scenario can be explored in further detail, without having to start the whole process again.
The European Urban Simulation for Asymmetric Scenarios project is funded by 20 nations under the Framework Program Force Protection of the European Defence Agency, a body that promotes the improvement of defense capabilities and developing better strategies for managing crisis situations in European countries.
Over the next year, Król and his colleagues intend to create more complex scenarios by adding cultural specific data to their models and recruiting more computing resources.
“During the next year we will prepare more complex scenarios with behavior modeling. A sample scenario regards controlling the access of pedestrians and vehicles to a military field installation during elections in a mission abroad. This type of scenario can have different variations, for example up to 20 civilians waiting in front of a camp entrance to an operation base. The military forces are tasked to ensure open entrances by calming and dispersing the crowd. Generally, the stress level on both sides is high, and a higher fear level can be observed in the local population. Direct contact with military personnel is generally stressful and there are high aggression levels on both sides,” he said.
The urban simulation project will finish in 2013 and could one day be used to save the lives of people on all sides.