Design Space Exploration of Ubiquitous Computing Applications using Agent-Based Simulations
|Presentation on||19. Jul 2011 17:00|
|Presentation room||Seminarraum I5|
|Add presentation to calendar||
This thesis investigates the use of agent-based simulations as a new method for exploring the design space of ubiquitous computing applications.
This approach is motivated by the challenges arising from the larger design space that characterizes ubicomp applications as a result of their distributed components. Agent-based simulations on the other hand provide a natural way to model the interactions between these components and to study the emerging effects of these interactions.
To enable simulations in which actors controlled by both agents and humans can interact, this thesis presents a light-weight approach for agent-based simulations, capable of execution at real-time, based on the well-established Belief-Desire-Intention agent modeling framework. Moreover, a concept for integrating this approach within the virtual simulation software FireSim is developed and an implementation is presented.
As a case study, an agent-based simulation was implemented for firefighters using a ubicomp system to support indoor navigation, called the LifeNet system, during search and rescue operations. The firefighter agent models were created based on the analysis by the author of empirical evidence obtained through two separate field studies of real firefighters using a LifeNet prototype. The LifeNet system is a good candidate for the evaluation of this method as it provides a relatively large design space. A set of design options was identified and modeled within the agent-based simulation for study in the context of a simulated search and rescue operation.
The evaluation of the simulated usage of the LifeNet System enabled the identification of a number of qualitative and quantitative characteristics of the relationship between the properties of the navigation support and its usage by firefighters on the one hand and global performance during search and rescue operations on the other. Some of these identified characteristics, such as certain trade-offs between the parameters involved are quite surprising and would have been difficult or impossible to anticipate prior to the simulations.
In summary, the presented study suggests that agent-based simulations provide a good trade-off between scalability for the exploration of ubicomp systems with large design spaces in terms of precisely defined research questions and the necessary effort to establish the required models grounded in empirical evidence.