My current research projects can be broadly classified into the following two areas:
· System Support for Distributed Simulation / Virtual Environments: A major project in this area is the “CoDS-GRID: Collaborative Large-scale Distributed Simulation on the Grid”, which will be described below. In addition to CoDS-GRID, I have two PhD students currently working on the following two problems: i) how the self-organization characteristics of the P2P overlay networks can be used to create an infrastructure to support distributed virtual environments; and ii) how an effective state update mechanism can be developed for distributed virtual environments, taking consideration of consistency and situation awareness. I am also a Co-PI of the project “Towards Highly Interactive Distributed Media Environment” supported by NRF. In this project, we aim to investigate some fundamental techniques in creating highly interactive, distributed media environments. Initial research result of this project has been published in DS-RT 2008.
· Modelling and Simulation of Large-Scale Complex Systems: A major project in this area is the “COSMOS: CrOwd Simulation for Military OperationS”, which will be described below. In addition to COSMOS, I am also involved as a Co-PI of the project “An Integrated and Adaptive Decision Support Framework for High-Tech Manufacturing and Service Networks” supported by A*STAR. The objective of this project is to investigate how design, analysis, enhancement and implementation of critical business processes in a manufacturing and service network could be realized using a single simulation/application framework. Under this project a generic symbiotic simulation framework has been developed and it has been applied to a number of applications. Research results of this project have been published in ICCS2008, PADS2008, and WSC2008.
CoDS-GRID: Collaborative Large-scale Distributed Simulation on the Grid
Modelling and Simulation (M&S) is an essential tool in many areas of science and engineering, for example, for analyzing natural phenomena or predicting the behaviour of new systems being designed. The development of such complex simulation applications usually requires collaborative effort from people with different domain knowledge and expertise, possibly at different locations. Furthermore, these simulation systems often require huge computing resources and the data sets required by the simulation may also be geographically distributed. This project will develop a generic framework for component-based model development and a service oriented simulation architecture for the Grid (see Figure 2). It consists of the following work-packages and is supported mainly by research students:
Figure 2: CoDS-GRID
· Component-based Modelling and Simulation: This focuses on two main problems: (i) composability of simulation components – developing mechanisms/methods to describe the semantics of simulation components, to select and assemble reusable simulation components in various combinations into simulation systems that meet the modeller’s functional requirements, and to validate the larger, integrated model; and (ii) interoperability of simulation components – engineering the integration of these smaller models, ensuring that the data exchange is mutually consistent with the interoperating simulation components’ semantics, and executing the integrated model seamlessly and efficiently. A framework has been developed by one of my Master students [PADS2006], which consists of schemas for collaborative development of simulation applications. One of my PhD students has also been working on developing a formal approach to specifying time synchronisation algorithms using Finite State Process (FSP) notation. This can be used for checking the composability of interoperating components, each of which has its own time synchronization mechanism.
· Service Oriented Architecture for Distributed Simulation: This is based extensively on our existing experiences of developing scalable frameworks for the execution of HLA-based distributed simulation on the Grid [PADS2006, DS-RT2005, DS-RT2002]. We have proposed a Service Oriented HLA Runtime Infrastructure (RTI) framework that implements RTI entirely using Grid services. Since the functionalities of an RTI are provided as Grid services, distributed simulations can be conducted without any vendor-specific RTI software. Other benefits of this service oriented implementation of the HLA, which is drastically different from the current HLA implementation, include i) modularity (a module can be re-implemented without affecting federation execution); ii) flexibility (various Grid services can be deployed and undeployed on demand); iii) fault-tolerance (fault-tolerance can be supported by replicating services); and iv) increased degree of interoperability and reusability (federates can be easily integrated into a simulation federation through these services). One of my PhD students is currently investigating issues on federate migration and fault-tolerance by taking advantages of this service oriented architecture [WSC2008, DS-RT2007].
· Collaboration Framework for Distributed Workflow Execution: There are multi-stages in the life cycle of modelling and simulation (M&S), from model development, execution, to validation and verification of simulation results. Specifically, in a collaborative environment, multiple parties may be involved at each stage, and thus bringing another dimension in the M&S life cycle. In a distributed collaborative environment, there is no centralized workflow engine that controls the sequence of activities. Instead, the specification of the activities and their sequence should flow through the modellers in a distributed manner. Similarly, to initialize the simulation execution, the specification of the composition of the simulation model (i.e., the composition of the simulation component services) should flow through the resources where the components are executed. Distributed orchestration also applies to the flow of work amongst various stages of the M&S life cycle so that members of a distributed design team and end-users with diverse expertise can collaboratively develop, document, discover, execute, test, analyze and share component-based simulation models. One of my PhD students has developed a collaboration framework using light-weight mobile agent and code-on-demand technologies to map the abstract job workflow specification to the dynamic Grid resources on the fly for distributed job workflow execution [JSA2008, JPDC2007, ICCS2007, ICDCS2005]. And one of my Master students is currently working on the problem to partition the workflow so that parts of the workflow can be executed collaboratively by multiple workflow engines.
COSMOS: CrOwd Simulation for Military OperationS
Crowd control becomes increasingly important in urbanized military operations such as peace keeping, riot control, disaster management, emergency evacuation, and rescue operations. Given the military challenges and risks imposed by the crowds, there is an urgent need to develop a system for military personnel to get prepared for handling various situations, to formulate strategies and answer “what-if” scenarios, and to evaluate hundreds of contingency plans so as to prioritize resources and time during an operation. One way to do so is to create a synthetic virtual environment and use Modelling & Simulation (M&S) techniques to emulate urbanized military operations. Crowd modelling and simulation is an essential component of such an environment. This project aims to develop a generic system for CrOwd Simulation for Military OperationS (COSMOS) with the integration of game AI, distributed simulation technologies, and computer graphics and animation.
Figure 3: COSMOS
This project is funded by Defense Science and Technology Agency (DSTA). As shown in Figure 3, it is a multi-disciplinary project that requires collaboration with colleagues in the School as well as collaborators from psychology and other engineering disciplines. The project is divided into the following five Research and Development Tasks (RDTs): RDT1 Behaviour Representation and Cognitive Models, RDT2 Ontology and Knowledge Repositories, RDT3 Agent-based Simulation Architecture, RDT4 Symbiotic Simulation, and RDT5 Animation and Visualization. The main research challenges are in the development of a realistic behaviour model and scalable agent-based simulation architecture to support large scale crowd simulations. Details of this project can be found from the project home-page: http://cosmos.ntu.edu.sg/.
For a realistic, large-scale crowd simulation, it requires: i) a cognitive model that takes consideration of social and psychological factors; ii) a smart representation of the virtual environment that makes reasoning and implementation of physical behaviour more effective; and iii) efficient mechanisms that support a large-scale agent-based simulation. Under this project, we have developed a cognitive model based on the social attachment theory in psychology and has applied the model to a scenario simulating emergency situations in an underground train station [CASA2008]. Currently, we are developing a new decision making framework, based on the Recognition-Primed Decision (RPD) model, to reflect “human-like” decision making process. The framework incorporates cognitive components, such as sensory system, attention, memory store, expectancies, course of actions, and goals. To support large-scale agent-based simulation, issues such as multi-resolution modelling and scenario-based partitioning, are also currently under investigation.