Please use this identifier to cite or link to this item:
Title: Tools for multi-agent crowd simulation : redesigning and extending the COSMOS crowd simulation suite
Authors: Nasri Othman
Keywords: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
DRNTU::Engineering::Computer science and engineering::Computing methodologies::Simulation and modeling
DRNTU::Engineering::Computer science and engineering::Computer applications::Social and behavioral sciences
DRNTU::Engineering::Computer science and engineering::Software::Software engineering
DRNTU::Engineering::Computer science and engineering::Software::Programming techniques
Issue Date: 2012
Abstract: Crowds are a highly complex social phenomenon and models to simulate crowds are highly difficult to develop. To tackle this difficulty, the COSMOS project was undertaken from 2006 to 2011 to investigate and develop a generic system for crowd modelling and simulation for military operations. In this final year project, we investigated the possibility of adapting the COSMOS suite into a modular multi-agent crowd simulation library which can be easily used by future crowd simulation models. This was accomplished by reviewing the existing COSMOS framework to identify problems, and then restructuring and extending it into a library. In addition, we analyzed the various simulation techniques used, resolving defects and optimizing performance as needed. A source code analysis of the existing COSMOS framework revealed that it was not well-coded, with a high degree of tight integration among the components. Less than 7% of the classes are abstract making it difficult for developers to extend the framework, and over half of the packages are involved in cyclic dependencies. Thus, the only possible way to improve the framework was to redesign it completely. Development consisted of three stages: design, restructure, and extend. A design philosophy was formulated to guide the development process. Essential components from the existing framework were restructured into over 8 major modular components. The new library was further extended with improvements and optimizations, including using faster and more scalable spatial indices for storing simulation entities. Over 350 classes were developed to construct the new library. Code analysis of the new library showed significant improvements over the existing framework. Over 27% of the classes are abstract, and no cyclic dependencies exist among the packages. By breaking down the system into modular components, the cyclometric complexity and average method length had also been reduced by over half. Due to the modular design, simulation models made in COSMOS can be readily executed on a variety of simulation platforms, including our own Prototyping Simulator, Repast Simphony and MASON, without any code modifications in the models. Our performance tests on the optimizations of key simulation components indicated substantial performance improvements of over two orders of magnitude under selected tests. In our scalability tests, we have also demonstrated that the new library is capable of efficiently simulating 20,000 agents without facing any significant bottlenecks, a feat that was not possible with the existing framework. With this new library, technology that was originally developed in the COSMOS project can now be easily used for future crowd simulation models. This includes novel behaviour modelling components such as our Recognition Primed Decision and Emotion Appraisal engines. The breadth of features supported by our library relieves modellers from having to implement the required functionality themselves. Lastly, the library is envisioned to be extensible. As crowd simulation is an emerging multi-disciplinary research field, we anticipate that the library will be continually extended with new crowd simulation components from future research.
Rights: Nanyang Technological University
Fulltext Permission: restricted
Fulltext Availability: With Fulltext
Appears in Collections:SCSE Student Reports (FYP/IA/PA/PI)

Files in This Item:
File Description SizeFormat 
  Restricted Access
6.43 MBAdobe PDFView/Open

Google ScholarTM


Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.