The INSTINCT (Innovative Science and Technology IN Counter-Terrorism) Programme is the overarching government initiative, led by a cross-government team anchored within the Office for Security and Counter-Terrorism (OSCT) in the Home Office.
INSTINCT team members come from more than a dozen different government departments and agencies led by the Home Office and involving the Ministry of Defence, their purpose is to introduce new scientific ideas into combating terrorism. In late November, AOS was selected to participate in the INSTINCT Technology Demonstrator Event in London, to demonstrate the capabilities of its behavioural modelling. Using our CoJACK/VBS2 system we modelled an urban crowd, complete with a policeman and his German Shepherd police dog.
Short movie clips of what we showed can be found here.
The Office for Security and Counter-Terrorism co-ordinates the INSTINCT programme within the Home Office and they’re seeking innovative solutions to address the objectives of the Government’s counter-terrorism strategy, known as CONTEST. INSTINCT’s key aim is to ensure that science and technology are fully harnessed to strengthen the UK’s ability to combat terrorism. The Technology Demonstrator Event showed innovative technologies that could help address counter terrorism problems in crowded places.
In response to the increased threat of terrorist attacks in public places, INSTINCT is seeking better understanding of crowd behaviour in congested areas. Areas like shopping centres, stadiums and transport hubs are potential targets and the goal is to reduce their vulnerability to terrorist attack. The Technology Demonstrator Event focussed specifically on technologies suited to modelling and predicting crowd behaviour, e.g., identifying abnormal crowd behaviour and detecting the causes of changes in crowd behaviour.
The scene below shows the policeman in the open space, surrounded by shoppers, business people, and two children. The children are playing with his dog. This broad range of characters illustrates the challenges of modelling civilians. There is a much wider diversity of goals, behaviours, behavioural cues and emotions among civilians than there are in the military scenes that have previously been the subject of immersive simulation systems, or serious games, such as VBS2.
AOS is breaking new ground with more accurate modelling of a wide range of civilian behaviours. In today’s military, serious games abound and are used to teach a variety of skills required by the warfighter, including tactics, protocol, procedures, mechanics and management. But current serious games have their shortcomings. Predictability, for example, where the computer entity (i.e. your ‘virtual actor’ adversary) executes the same task (e.g. ducking for cover) in the same way over and over. Another is variance – there’s usually no difference between actors and there’s certainly no variability within an actor. These may sound insignificant in a computer game but they’re a major break with reality. Humans vary not only among each other (i.e. you may do things differently than someone else) but each individual also varies across time (i.e. you may approach the same challenge in different ways depending on your current mental or physical state). When these important human characteristics aren’t modelled in a serious game environment, realism suffers significantly along with the effectiveness of the game as a training device.
So while the need for realistic behaviour is recognised, the problem is how to model variance and reduce predictability in serious games. Fortunately, human behaviour representations are becoming increasingly rich, building on data from the cognitive and affective sciences. Modelling these results would produce more realistic behaviour and resolve the issue, but this introduces another problem. How? In fact, several approaches to modelling human behaviour are available but most of these cognitive architectures are difficult to use and time consuming to develop.
But AOS’s CoJACK, the only COTS system based on the BDI (Beliefs/Desires/Intentions) paradigm, is leading the way toward more realistic behaviour.
Using this BDI representation gives CoJACK several distinct advantages over other approaches. While other cognitive architectures, e.g. ACT-R, typically represent procedural knowledge in terms of fine-grained steps, CoJACK offers an easy-to-use high-level representation underpinned by sub-symbolic computations that influence processing without obscuring the high-level viewpoint.
Today’s serious games are often built on video game engines that provide the human player with impressive photo-realistic 3D environments and believable-looking virtual actors/adversaries. But in order to meet the sophisticated expectations of the current generation of gamers, behavioural realism is an absolute requirement in an effective modelling of civilian behaviour and crowds. When virtual actors behave in an intelligent, variable but repeatable fashion, the resulting scenarios will provide both absorbing content and truly challenging situations that can closely mimic real life.