Modeling Human Behavior in CHAOS
|has title::Modeling Human Behavior in CHAOS|
|Master:||project within::Cognitive Science|
|Student name:||student name::Stefan Boronea|
|Second reader:||has second reader::Emiel Ubink|
TNO develops and applies simulations as a research tool in order to assist its customers, such as the Dutch military and the Dutch fire department. These simulations require accurate models of human behaviour and performance. In these models, many factors are addressed, from equipment to environment and from physiology to psychology. With such a large number of variables, a rule-based approach quickly becomes incomprehensible. TNO has therefore taken another approach with the CHAOS behaviour architecture.
CHAOS (Capability-based Human performance Architecture for Operational Simulation) is a pluggable behaviour architecture, meaning that different behaviours and stressors can be added to, or removed from the system, without altering the integrity of the system as a whole.
CHAOS consists of “demons” that are “shrieking” for attention in an arena called the pandemonium. These demons represent either behaviours (actions) or stressors. Besides demons, CHAOS also contains capabilities that play a central role in the performance modelling and action selection mechanisms.
The implementation behind the CHAOS architecture is based on common sense and pragmatic considerations. Although its testing has been successful, a solid scientific base for the model is lacking. The previous tests primarily examine the effects of stress and/or multi-tasking on human performance.
The goal of this research project is to take the validation of CHAOS a step further by implementing the empirical findings in the CHAOS architecture. In order to achieve this, human behaviour models for various scenarios will be developed and validated within the context of the examined tasks, including the definition and formalization of task-relevant resources. Resources and performance models will then be combined into a human behaviour model that is constructed in Java, using the CHAOS library.
Abstract KIM 1
The project is set on tackling human behaviour from a high-level design perspective, while at the same time trying to integrate existing research on human behaviour modelling. In principle, human behaviour may be considered as either purely reactive, in which physical response holds the greatest interest, or otherwise considered as having a broader dimension that includes some aspects related to top-down cognitive biasing and attentional behaviour.
The current study focuses primarily on how to model the effects of anxiety on the performance of subjects in a dual-task setting. In this sense a CHAOS model needs to be designed and its output compared to the data obtained in two different experiments. The first experiment, which I will discuss in more detail during the meeting, tries to determine the influence of anxiety on a running and aiming task, where anxiety is induced by increasing the height at which the subject must run. From the model we would like to determine a number of functional components which allow this effect to be seen and additionally validate it against the attentional control theory (Eysenck et al., 2007).
With this we would also like to see how the CHAOS modelling framework is adapted to the previously mentioned settings, aiming at either a scientific validation for this specific pandemonium-based approach or at a redesign of the framework and modelling approach.
Abstract KIM 2
Human behaviour fascinates the minds of scientists and non-scientists alike, reaching beyond the borders of numerous domains, stemming from the philosophical and psychological aspects and growing in the past several decades towards more pragmatic applications in technology and artificial intelligence.
The increased interest in creating realistic, biologically plausible models of behaviour has brought forth specialised paradigms and frameworks that help formalise, functionally decompose and integrate complex behaviour. These formal descriptions of behaviour require however a number of experimentally attainable heuristics, such as internal parameters, statistical inference models, states, transitions and interactions.
This thesis takes a look at the process behind the creation of human behaviour models based on experimental data, with an emphasis on one particular framework, the CHAOS modelling framework. The outcomes of the modelling process are then discussed, providing a number of experimental and framework improvements that could help facilitate the design and implementation of future modelling frameworks and human behaviour models.