AOS at SimTecT 2010, Brisbane

Screen shot 2010-06-02 at 9.24.28 AM1st June: Dr Rick Evertsz presented the paper "Realistic Virtual Actors for Training in Counterterrorism" at the SimTecT 2010 Conference, Brisbane Convention Centre. The presentation reported on how realistic virtual actors can be used to provide the counterterrorism trainee with experience in detecting abnormal behaviour and using tactics that increase the visibility of a terrorist concealed in a crowd.

Terrorism has become a common international threat, and is particularly problematic to counter because it occurs in a civilian rather than military context. It is difficult to differentiate terrorists from normal civilians because their strategy is to merge into the crowd and thereby avoid detection. Providing security personnel with the necessary experience to spot terrorists in a crowd can be time consuming and resource intensive. Simulation can provide a valuable complement to current approaches, making them more cost effective by reducing the manpower and infrastructure required to support training.

Our approach is to use the cognitive architecture, CoJACK™, to create virtual actors (civilians, terrorists) that react variably but realistically to the trainee’s and other virtual actors’ actions. These virtual actors are imbued with emotions, such as fear, anger and anxiety, which lead them to react plausibly to events in the scenario. The key point is that these reactions are not scripted but emerge from the interplay between the virtual actor’s cognitive and emotional faculties. This yields a flexible training environment that is responsive to what the trainee does. In many ways, it is more difficult to generate realistic behaviour in a civilian rather than military context - the actions and reactions can be quite subtle. The presentation outlined a civilian counterterrorism scenario, in which there is a terrorist on reconnaissance, and diverse types of civilian with varying goals and emotional states. The virtual environment, VBS2™ was used to represent the physical environment and the characters’ embodiments.

Screen shot 2010-06-02 at 9.26.18 AMA scenario was specified with the objective of providing a focus for understanding and modelling crowd behaviour in congested areas and to provide a case study where terrorist behaviour could be distinguished from that of civilians. Areas like shopping centres, stadiums and transport hubs are potential targets and so this scenario was set in a pedestrian precinct.

The trainee controls the policeman. The scenario was populated with a mix of about 50 virtual actors:

  • Workers (roughly 40%)
  • Tourists (roughly 20%)
  • Shoppers (roughly 40%)
  • Children (roughly 1%)
  • Mother/father (roughly 1%)
  • Reconnaissance Terrorist (1 instance)
  • Police dog (1 instance)

Each individual virtual actor is assigned a “personality” assembled from a palette of capabilities. A personality consists of the following main components:

  • Beliefs – what the virtual actor knows (e.g. location of particular shops, where the bus stop is, who its family group is, etc.).
  • Procedures – what the virtual actor knows how to do (e.g. how to navigate to a given location, how to respond aggressively when obstructed, etc.).
  • Goals – what the virtual actor wants to achieve (e.g. go shopping, get to work on time, etc.).
  • Temperament – the affective style of the virtual actor (e.g. naturally anxious).
  • Mood – a diffuse affective state of relatively long duration, (e.g. being in a state of fear due to some event).
  • Attitudes – predispositions to objects or situations (e.g. not liking police dogs, which types of event are considered dangerous, etc.).
  • Cognitive Effectiveness – the baseline state of the underlying cognitive parameters, affecting capabilities such as ability to recall facts, hold intermediate results in working memory, and stay focused on a goal.

The virtual actor’s personality determines how it responds to events in the scenario and what actions it chooses to perform. For example, if a businessman’s goal is to get to work on time and he knows the way to work (beliefs) and knows how to navigate to that location (procedures), he will walk to work. However, if he is late (belief), he will run (procedures). If he is obstructed by a policeman, and is ill tempered (temperament) or in a bad mood, he may respond aggressively (procedures). If he is fatigued he may forget to look both ways when crossing a road (cognitive effectiveness).

The virtual actors can be assigned personalities manually, or automatically according to a predetermined probability distribution. Once this is done, the scenario can be started and it plays out in a way that is influenced by the interaction between the trainee’s actions and the personalities of the virtual actors. The behaviour of the virtual actors emerges from these interactions. It is not scripted to produce a particular outcome. In this way, the trainee gets to experience a more variable interaction than is possible with typical scripted approaches to implementing human behaviour.

If desired, the trainer can bias the scenario in a particular direction to focus the development of the trainee’s skill set. An example would be to turn up the baseline level of fear for all civilians. This could be used, for example, to simulate a situation in which there is a heightened terror alert and the general populous is a bit jumpy. Moderators on individuals can be tweaked at runtime to trigger behaviour that is apposite to the desired training outcome. For example, lowering the fearfulness of the reconnaissance terrorist so that he is less affected by the trainee’s actions.


Acknowledgements

The development of CoJACK was funded by the IHBR (Improved Human Behaviour Representation) project, funded by DAES (Directorate of Analysis, Experimentation and Simulation) UK Ministry of Defence.

We thank Bohemia Interactive for their support in creating the London street scene and UK-specific characters.

We are also grateful to Detica for their support during the development of the demonstration.

Privacy Policy | Contact Us | Site Map                 © AOS Group 2015