School of Computer Science and Information Technology

Agents at RMIT

NEW Postdoc position now available - closing 5th of June 2008 NEW

Positions Vacant

The Computer Science Department at RMIT has research in a wide range of areas relevant to Intelligent Agents. We describe here the main areas in which we are working and the specific sub-projects within each area. The agent group is part of the "Intelligent Systems" area within the School of Computer Science and Information Technology. Other research which may be of interest to aspects of intelligent agents can be found on the web pages of other research groups within the School, accessible from the research areas page. We are very interested in hosting visiting researchers and periodically have postdoc positions available. Please contact Lin Padgham if you are interested in visiting our group.

This site covers the following topics:

Members

Staff:

PhD students

  • David Poutakidis
    Debugging Multi-Agent BDI Systems
    Supervisor: Lin Padgham, second supervisor: Michael Winikoff
  • Lavindra de Silva
    Planning in BDI Agent Systems
    Supervisor: Lin Padgham, second supervisor: Michael Winikoff
  • Chris Cheong
    A New Approach To Agent Interactions: Towards Flexibility And Robustness
    Supervisors Michael Winikoff and Lin Padgham
  • Simon Duff
    Enhancing Goal Selection for Intelligent Agents
    Supervisors: James Harland and John Thangarajah
  • Khanh Hoa Dam
    Supporting Software Evolution in Agent Systems
    Supervisors: Michael Winikoff and Lin Padgham
  • Jennifer Sandercock
    Using Emotions and Learning to Improve Adaptation and Individuality of Non-Player Characters
    Supervisors: Lin Padgham and Fabio Zambetta
  • Carlos Alexandre Queiroz
    Topic: Distributed Situation Awareness
    Supervisors: Lin Padgham
  • Amir Aryani
    Topic: TBC
    Supervisors: Michael Winikoff and Ian Peake
  • Dhirendra Singh
    Topic: Agent Learning for Energy Management
    Supervisors: Lin Padgham
  • Zhiyong Zhang
    Automated Test with AOSE Methodology
    Supervisors: Lin Padgham and John Thangarajah

Areas of Work

The group has work in 3 main areas: Agent Oriented Software Engineering, Agent Reasoning, and Open Agent Systems, funded by a variety of grants.

(1) Agent Oriented Software Engineering

This area is funded by the grant Simplifying the Development of Agent Oriented Systems and its successor Advanced Software Engineering Support for Intelligent Agent Systems.

Prometheus Methodology

Prometheus is a software engineering methodology that supports the engineering of agent-oriented software. (more information)

Current Participants: Lin Padgham, Michael Winikoff, John Thangarajah, Khanh Hoa Dam

Previous Participant: Shankar Srikantaiah

This work is being done in collaboration with Agent Oriented Software Pty. Ltd.

Debugging and Testing

Current Participants: David Poutakidis, Lin Padgham, Michael Winikoff, Zhiyong Zhang

Component Based Agent workbench

Current Participants: Lin Padgham, Michael Winikoff

Previous Participant: Gaya Jayatilleke

(2) Agent Reasoning

Planning and Learning in Intelligent Agents

This area is funded by the grant Planning and Learning in BDI Agents BDI agents do not create plans in the traditional planning sense, and neither are they able to learn from experience. We are investigating extending BDI agents in these directions. This work also includes development of a formal operational semantics for a goal based BDI language which is integrated with an HTN functionality.

Current Participants: Sebastian Sardiña, Lavindra de Silva, Lin Padgham, Michael Winikoff

Previous Participant: Toan Phung

Logical Agents

Agent systems require reasoning capabilities, as well as a programming environment. Hence one approach to the development of agent systems is to start with the paradigm of logic programming, which provides both reasoning and programming facilities, and to extend and modify it to incorporate the various features of agent systems, such as reactivity, autonomy, social abilities and so forth. We have identified a method for integrating resolution-based methods of goal search with forward-chaining methods in linear logic, thus providing a smooth integration of rationality (or backward-chaining) with reactivity (or forward-chaining). A prototype system based on this integration is underway.

Current Participants: James Harland, Simon Duff,

Previous Participants: Abdullah-Al Amin, Duc Quang Pham, Michael Winikoff

See also the Lygon homepage

(3) Open Agent Systems

This area is funded by the grants: Open agent architectures for intelligent distributed decision making and Description, Composition, Discovery and Deployment of Intelligent Agent Services Application areas include: Current Participants: Lin Padgham, Michael Winikoff, James Harland

Previous Participants: Ian Mathieson, Bao Quoc Vo, Antony Iorio, Kenichi Yoshimura, Wei Liu, Tom Gamble, Aloys Mbala, Min Xu.

Composing Agent Services

Current Participants: Lin Padgham, Wei Liu

Previous Participant: Bao Quoc Vo

Teamwork

One of the important issues in multi-agent systems is how to get agents to co-operate with each to achieve things they cannot achieve alone. Often, although not always, this cooperation will be in the context of a team or subteam of agents which agree to work together. This project is focussed on mechanisms for flexible and efficient cooperation between agents in dynamic environments. A significant amount of the work on this project has been in the context of Robocup, both the simulator league and the large robot league.

Current Participants: Lin Padgham, Michael Winikoff.

Previous Participants: Joshua Hutchison, Simon Ch'ng, Malcolm Robins,David Poutakidis, Christopher Ho Mok Cheong (Honours 2003)

Grants

A Framework for Adaptive Extensible Personae for Interactive Toys

ARC Linkage Grant (2008-2010, grant LP0882013)

This project will be of benefit in that it will facilitate the development of quality toys, based on sound psychological foundations. These toys will have a long life since they can be extended over time, growing with the child. There is also potential to be used for children with psychological difficulties, and applications in health and aged-care. This project will provide opportunities for Australian business, and will also help to keep Australia at the forefront in the area of intelligent agent technology. It is an example of a project that helps create a culture of innovation in Australian industry.

Current Participants: Lin Padgham, Michael Winikoff, Lawrence Cavedon, Barbara Kelly, Fabio Zambetta, RB Wesson,
Partner Organisation: XSiVE Pty Ltd.

Service-oriented negotiation and coordination in multi-agent systems

ARC Discovery Grant (2006-2009, grant DP0663147)

There is an increasing trend towards structuring software as a collection of autonomous entities that negotiate and coordinate. We propose to use the concept of Best Alternative To a Negotiated Agreement (BATNA), proposed by management and political scientists, as a basis for negotiation. Expected outcomes from this work include a formalisation of BATNA, and a flexible agent negotiation and coordination framework that allows agents to use negotiation to deal with coordination breakdowns. An advantage of our proposed approach is that it avoids unrealistic assumptions, such as requiring all agents to be perfectly rational, or requiring excessive amounts of common knowledge.

Current Participants: Lin Padgham, James Harland, Michael Winikoff, Bao Quoc Vo

Acts of electronic negotiation: Overcoming communication barriers to transdisciplinary innovation in design

ARC Discovery Grant (2006-2008, grant DP0665744)

Improved collaboration between the diverse contributors to design in the construction industry is urgently needed to reduce waste, streamline production and improve performance. It is also a key to better and more advanced design. The quality of human to human interaction in computer mediated environments is critical to the occurrence of innovation between design disciplines. This project will create a pool of highly qualified personnel in this area in Australia, including participating designers introduced to novel empowering approaches to network communication. It will develop and apply knowledge from other disciplines to developing tools for the design community.

Current Participants: Mark Burry, Lin Padgham, Mike Xie, Andrew Burrow

Planning and Learning in BDI Agents

ARC Linkage grant (2005-2007) LP05... with Agent Oriented Software Pty. Ltd.

Advanced Software Engineering Support for Intelligent Agent Systems

This project is funded by an ARC Linkage grant (2004-2006, grant LP0453486) with Agent Oriented Software Pty. Ltd.

Software Agents are an important technology for developing the complex software systems that are increasingly required to meet the needs of society. A crucial obstacle to the widespread adoption of agent technology is the lack of an appropriate software engineering methodology. This project proposes to explore support for design processes addressing advanced issues in agent systems, such as goal-based requirements, debugging using design artefacts, component-based design, and reuse. We will also extend the methodology to support teamwork and open systems. We will be building on successful work we have already done in establishing a basic agent oriented software design methodology.

This work builds on work done as part of the project Simplifying the Development of Agent Oriented Systems.

Current Participants: Lin Padgham, Michael Winikoff, Andrew Lucas (Agent Oriented Software), Andrew Hodgson (Agent Oriented Software).

Previous Participants: Liz Haywood

Industry Partners: Agent Oriented Software Pty. Ltd.

Description, Composition, Discovery and Deployment of Intelligent Agent Services

This project is funded by a DEST IAP grant (2004-2006, grant CG040014). This project is participating in Satine (Semantic Based Interoperability Infrastructure for Integrating Web Service Platforms on Peer to Peer networks), a European Union 6th framework project (IST-1-002104-STP). Descriptions of some of the web services developed can be found at our testbed site.

The World Wide Web is currently a major source of information for individuals, businesses and the public sector. The vision for future use of the WWW and the Internet, is that in addition to being a repository for static and dynamic informational pages, the Internet will also host a range of intelligent agents, offering services to both human users and to other intelligent agents. This will be an open and evolving system, where as new agents are added, other agents are able to locate them and to avail themselves of their services as appropriate. In order for this vision to be fully realised there are a wide range of technical issues which must be addressed. We will be working with international and Australian partners to explore issues in the areas of flexible and scalable distributed directory services, service description languages, and service composition and execution, while also developing and deploying agents in the area of tourism.

Current Participants:
Main RMIT participants: Lin Padgham, Ian Mathieson
Other RMIT participants: Michael Winikoff, James Harland

Industry Partners: Tourism Victoria, Australian Tourism Data Warehouse

Other Partners: Satine (EU 6th framework project), openNet,
Wei Liu (University of Western Australia)

For more information see the DEST grant homepage and the internal project homepage (restricted).

Open agent architectures for intelligent distributed decision making

This project is funded by an ARC Linkage grant (2003-2005, grant LP0347025) with the Bureau of Meteorology and Agent Oriented Software Pty. Ltd.

Sophisticated software systems are part of the essential infrastructure of our daily lives. Complex systems such as the internet, finance, or telecommunications software cannot have a centralised point of control or a single developer. The aim of this project is to develop an architecture and support infrastructure enabling intelligent agents to locate and use services in such open systems. The fundamental questions that must be addressed include issues such as how agents will find and use newly added services and how services will communicate with each other, given that they are developed independently.

Current Participants: Lin Padgham, Michael Winikoff, Sandy Dance (Bureau of Meteorology), Ralph Rönnquist (Agent Oriented Software)

Previous Participants: Ian Mathieson, Aloys Mbala.

Industry Partners: Bureau of Meteorology Research Centre and Agent Oriented Software Pty. Ltd.

For more information see the BMRC linkage grant homepage and the internal project homepage (restricted).

Presentation of Complex Systems

This project is funded by an ARC Discovery grant (2003-2005, grant DP0346691)

This project aims to enhance the understanding of multidimensional and multisensorial presentation environments. It will do this by developing a presentation system, driven by database information controlled by intelligent agent software, activating a dynamic visual and sonological environment, which is then evaluated in a variety of scenarios.

Current Participants: Mark Burry (Architecture and Design), Greg More (Architecture and Design), Lin Padgham, Ian Mathieson.

Previous Participants: Aman Sahani.

For more information see the internal project homepage (restricted).

High Energy Physics: GRID

This project is funded by VPAC Expertise grants (EPPNRM108.2003 and EPPNRM121.2004) in conjunction with the Experimental Particle Physics (EPP) group at The University of Melbourne as part of the HEPGrid project (EPPNME091.2003).

The analysis of HEP/EPP experiments such as BELLE and ATLAS involves complex filtering of vast amounts (potentially terabytes) of distributed and replicated data. Key problems in managing such DataGrid computations include locating appropriate datasets for the analysis, then scheduling and monitoring the progess of the distributed computations, where even the middleware support is experimental (and unreliable). This project explores the application of software agents to HEPGrid computations.

Current Participants: Lin Padgham, Ian Mathieson.

Partners: University of Melbourne, Physics, University of Melbourne, Computer Science.

Previous Participants: Tom Gamble, Antony Iorio, Wei Liu, Aman Sahani.

For more information see the HEPGrid Agent homepage. The internal project homepage is restricted.

Flexible and Robust Protocol-Based Interaction between Agents in Open Systems

This project is funded by an ARC Linkage grant (2002-2004, grant LP0218928) with Agent Oriented Software Pty. Ltd.

There has been much discussion of the importance of software agents for supporting a wide variety of interaction between businesses and individuals over the internet. Important applications include ecommerce and b2b applications. For the potential of software agents to be realised in open systems, issues of flexibility, robustness, and extensibility are critical. This project addresses the development of flexible and powerful mechanisms for interaction, within the context of FIPA (Foundation for Intelligent Physical Agents) standards. The project uses FIPA's Agentcities as a test bed, enabling us to build on and co-operate with a large European project starting mid 2001.

Current Participants: Lin Padgham, James Harland.

Previous Participants: Wei Liu, Kenichi Yoshimura, Aloys Mbala. Min Xu.

Industry Partner: Agent Oriented Software Pty. Ltd.

For more information see the protocols homepage

Simplifying the Development of Agent Oriented Systems

This project is funded by an ARC SPIRT grant (2001-2003, grant CO0106934) with Agent Oriented Software Pty. Ltd.

Design and programming of agents is a significantly different paradigm to either object-oriented programming or procedural programming. In this project we are investigating the key concepts necessary for effective agent design and implementation. We are developing a semantics suitable for understanding by an average programmer as well as development methodologies and tools. An initial methodology suitable for use with current BDI systems has been developed, along with supporting design templates and structures. This has been used in a number of workshop courses teaching agent-oriented design.

Current Participants: Lin Padgham, James Harland, Michael Winikoff, David Poutakidis, John Thangarajah

Previous Participants: Anna Edberg, Christian Andersson.

Industry Partner: Agent Oriented Software Pty. Ltd.

For more information see the SAC homepage



Publications

The following lists some recent publications. in reverse chronological order (i.e. most recent first). A full list is under construction and will be made available when it is done. In the meanwhile, papers can be found at their author's homepage.

Courses

CS435 Intelligent Agents and Agent Oriented Systems

This is an elective subject for the honours year in Computer Science and is part of the Intelligent Systems cluster for the coursework Masters degrees in the Department of Computer Science.

The subject aims to give an understanding of the research area of intelligent agents by means of reading and discussing a selection of research papers as well as briefly investigating at least one implemented system for agent programming.

The prerequisites for this subject are an introductory understanding of Artificial Intelligence techniques, particularly the area of planning, and an understanding of first order logic sufficient for understanding formal descriptions of extensions to FOL.

This subject will be run in a seminar mode where articles are discussed and analysed. In addition to developing knowledge of the content area students will develop skills in critical reading of research literature and in synthesizing and comparing approaches to problems.

Students will be expected to participate actively in the discussions, and to take it in turn to lead the discussions. Discussion leadership will involve preparation of focus questions as well as leading of the discussion in class.

CS436 Intelligent Web Systems

This subject introduces a wide range of Artificial Intelligence techniques which are used on the www. It is not limited to agent technology but does cover several types of web agents such as chatterbots, intelligent search agents and mobile agents. It focuses on the AI technology used in these systems.

CS526 Agent Oriented Programming and Design

This subject covers design and programming in the emerging field of agent-oriented software. The high level design methodology covered is appropriate for a wide range of approaches to agent development. The subject covers identification of agent roles, agent data sources, role interactions, use cases, agent types and agent interaction diagrams. The detailed design and programming focusses on the "Belief, Desire, Intention" framework which is particularly suitable for intelligent agents and has a strong industry base in Melbourne. Programming is done using JACK, a commercial, java based agent development environment. Agent oriented systems are a growing area and the field of agent oriented software engineering is in its infancy.

Previous Projects

Conflicting Goals and Intentions in Intelligent Agents

BDI agents are goal-oriented and will pursue their goals via whatever plans are available, looking for new plans if a particular plan fails. An agent can be pursuing multiple goals in parallel and actions towards these goals may be interleaved. This leads to situations where it is possible for an agent to act in ways that interfere with its own intended outcomes, without being aware of the situation. This project is exploring ways for an agent to be able to detect and resolve internal conflict situations in order to exhibit internally coherent behaviour.

Participants: John Thangarajah Lin Padgham, Michael Winikoff

Observation and Expectation

Observations and expectations are critical concepts in guiding intelligent agent behaviour. We are working on a formalisation of intelligent agents where these concepts are central and will also be developing a prototype system.

Participants: James Harland, Vu Binh Tran

Agent Cities

Agent Cities is a project run by FIPA, partly to use and test the agents standards which they are developing. The aims of this research project are to explore issues of interoperability between agents, including the adequacy of the infrastructure and the protocols developed. We are currently working on development of an agent to contribute to this site, as well as an agent which can interact with other agents on the site. The agent cities site can be found at http://www.agentcities.org

Participants: Lin Padgham, Kenichi Yoshimura, Keith Villiers Lily Sun, Antony Iorio, Nathaniel Orchis, Wei Liu, Jason Khallouf

Robocup F2000 leagues

We participated in the large robot league of robocup in 1998, 1999 and 2000, in co-operation with the Computer Systems Engineering Department who built the robots. This work has identified a large number of issues relevant particularly to physical systems. We have used a combination of behaviour based architecture for low level behaviours, with a BDI architecture for deciding which behaviours to employ. In 2000 we succeeded in getting into the quarter finals, playing against the champion teams from 1998 and 1999.

We are not currently participating in the F2000 league of Robocup, but are continuing some robotic projects with small robots.

Participants: Lin Padgham, James Brusey, Gregor Hall Malcolm Robins, Jey Srikantha, Vasu Uppu, David Poutakidis, Steven Garcia, Naree Song, Joseph Antony.

Robocup Rescue

Robocup Rescue is a new initiative started by the Robocup federation. The aim is to promote research in multi-agent systems dealing with issues of disaster management and recovery. The current simulator system is a high fidelity graphical simulation of a portion of Kobe during the 1996 earthquake. This work focusses particular on investigation of co-operation between heterogeneous agents, possibly built by different groups. There is also a strong focus on uncertainty and incomplete information. This is a new project which we are becoming involved in.

Participants: Lin Padgham, James Harland, Mohit Gupta,

Agent Architectures

This work focusses primarily around issues of when it is preferable to use a behaviour based architecture and when an architecture based on BDI systems, and in particular how to effectively integrate the two in physical systems which require low level behaviours. It also includes issues regarding the architecture of the BDI system and has included work on integration of an emotion component into the agent architecture. Although most of the work is in the context of multi-agent systems, architectural issues have so far focussed primarily on the within-agent architecture.

Participants: Jennifer Kashmirian, Christian Guttman, Michael Heemskerk, Ben Fortuna, Guy Taylor, Anthony Kendall, Lin Padgham, Gregor Hall, James Brusey

Some past students

  • Vu Binh Tran
    Expectation Logic
    Supervisors: James Harland and Margaret Hamilton
    Thesis passed October 2005.
  • John Thangarajah
    Managing Multiple Concurrent Goals in Intelligent Agents
    Supervisor: Lin Padgham, second supervisor: Michael Winikoff
    John's thesis was passed in mid-2005. He is working with the agent group as a post-doc.
  • Mikhail Perepletchikov (Honours 2004)
    Supervisor: Lin Padgham, second supervisor: Liz Haywood
  • Jason Khallouf (Honours 2004)
    Supervisor: Michael Winikoff, second supervisor: Liz Haywood
  • Robert Tanaman (Honours 2004)
    Supervisor: Michael Winikoff, second supervisor: Keith Frampton
  • James Brusey
    James finished his PhD in late 2002 and is now working in at Cambridge University in the Engineering Department's Institute for Manufacturing.
    http://www-mmd.eng.cam.ac.uk/people/jpb54/jpb54.htm
  • Arran Bartish (Honours 2001)
    A comparative Analysis of Intelligent Agent's and State Machines: Models for the game domain
    Supervisor: Charles Thevathayan, second supervisor: Peter Bertok
  • Khanh Hoa Dam (Masters, 2002-2003)
    Evaluating Agent-Oriented Software Engineering Methodologies
    Supervisor: Michael Winikoff, second supervisor: Lin Padgham
  • Joshua Hutchison (Honours 2002)
    Evaluating the JACK Teams Paradigm for Robocup Rescue
    Supervisor: Lin Padgham, second supervisor: Michael Winikoff
  • Mohit Gupta (Honours 2002)
    Building strategies in Roborescue
    Supervisor: James Harland
  • Duncan Bayly (Honours 2001)
    Supervisor: James Harland
  • Keith Villiers (AgentCities project 2003)
  • Hiten Ravani (summer student 2003)
  • Sheilina Geerdharry (summer student 2003)
  • Sindawati Hoetomo (summer student 2003)
  • Yenty Frily (summer student 2003)
  • Nathaniel Orchis (summer student 2003)
  • Chandaka Fernando (project student 2001)
  • David Charlton (SE agent project 2001)
  • Michael Styles (SE agent project 2001)
  • Michael Wong (SE agent project 2001)
  • Mukesh Chawla (summer student 2002)
  • Lito Cruz (PhD student)
    Supervisor: Lin Padgham
  • Rosette Kidwani
    Supervisors: Lin Padgham and Michael Winikoff
  • Gaya Jayatilleke
    A Framework for Component Based Agent Design
    Supervisors: Lin Padgham and Michael Winikoff
    Gaya submitted in early 2007 and his thesis was passed later that year.
  • Toan Phuoc Phung
    Supervisors: Michael Winikoff and Lin Padgham
    Toan submitted early in 2007 and his thesis was passed later that year.
  • Quang Duc Pham
    A logic-based approach to flexible agent interaction protocols
    Supervisors: James Harland and Michael Winikoff
    Duc submitted mid-2007 and his thesis is currently under examination.

Some former staff

Notes