Dr Andrew Coles

Introduction

I am a SICSA Research Fellow based at the University of Strathclyde, working within the Strathclyde Planning Group. SICSA is a collaboration between Scottish Universities to further develop Scotland's international standing in Computer Science research. Within this framework, I will be conducting outwards-facing research in my own field — AI Planning — with the aim being to foster collaborations with other Scottish universities. As part of this, I am a member of THREADSS, and have been working with Stephen Gilmore on how AI Planning can be used to make configuration choices for PEPA models. I am also keen to hear from anyone who is interested in what planning might have to offer; or indeed what they may have to offer planning.

Prior to my current position, I was involved in an EPSRC project investigating planning with continuous numeric resources. This project is still ongoing and already has a number of successful outputs including the planners CRIKEY3 and COLIN.

Research

COLIN

For information on COLIN, the first PDDL planner to support (linear) continuous numeric change click on the following link: COLIN Homepage. An application of COLIN to scheduling aeroplane landings has recently been featured on the University news website.

CRIKEY 3

CRIKEY 3 is a planner that supports the full temporal semantics of PDDL 2.1. Visit the CRIKEY Homepage for more information.

Upwards

The optimal planner Upwards competed in the 2008 International planning competition. Note that Upwards considers an optimal plan to be one containing the smallest possible number of actions and is not 'cost optimal'. For information on Upwards and to download the source code see the IPC 2008 planners website.

TSGP

For information on TSGP, a decomposition based planner, click on the following link: TSGP Homepage.

Local Search and Generic Types in Forward-Chaining Planning

From 2003 to 2006 I studied for a PhD in computer science at the University of Strathclyde. In 2006 I submitted my thesis, concerned with applying local search techniques to forward-chaining planning; and the use of generic type information to improve the relaxed planning graph (RPG) heuristic, as used in FF.
In my thesis the use of various local-search algorithms and neighbourhood functions when performing forward-chaining planning are investigated, and to reduce the impact of the run-time distribution under differing random seeds, a multi-point strategy with random restarts has been implemented.
To provide further heuristic guidance, an approach to improving the feasibility of relaxed plans through adding additional actions which can be determined using generic type information has been implemented; in doing so, the identification of generic types can lead to extra search guidance without relying on being able to cleanly seperate the induced subproblem, as needed in previous approaches.
A link to a PDF of my thesis is given at the bottom of my list of publications.

Pac-Plan — Probabilistic Planning with a Deterministic Numeric Planner

Amanda Smith and I have an on-going collaboration with Sergio Jiminez, Universidad Carlos III de Madrid, developing a probabilistic planner — Pac-Plan — based on replanning using a deterministic planner without discarding the probabilities of actions' outcomes. Click here to download Pac-Plan, or for a workshop paper on the work as of December 2006 see my list of publications.

Marvin

For information on the planner, Marvin, which competed in the Fourth International Planning Competition, see the the Marvin home-page. For the JAIR article on Marvin, see my list of publications.

Earlier Work

Work in the early stages of my postgraduate studies was concerned with managing resource constaints in planning, Amanda Smith and I produced an implementation of an algoritm for calculating resource envelopes incrementally. The implementation is that of an algorithm devised by T. K. Satish Kumar; his paper, entitled 'Incremental Computation of Resource-Envelopes in Producer-Consumer Models', which describes the algorithm can be found on his website. The C++ source code of the implementation is available here and is known to compile using g++ version 3.3.2 running on Linux. The slides for a talk we did describing the algorithm can be found here.

Before starting at Strathclyde I gained my first degree in Artificial Intelligence at the University of Durham. For my final-year project I developed a simulation of rovers operating within a closed Martian environment. A domain, based on the Rover domain from the Third International Planning Competition was produced which enabled the integration of a number of planners; notably Metric-FF, MIPS, Sapa and VHPOP.

The source code of the simulation is available on request; my project report can be found here.

Publications

View my list of publications here.

Teaching

Click here for a link to information about my teaching activities

Professional Activities

Programme Committee Memberships

• 21st International Joint Conference on Artificial Intelligence (IJCAI 09)
• 27th Workshop of the UK Planning and Scheduling Special Interest Group (PlanSIG 08)

Workshop Organisation

I am a co-organiser of the ICAPS 2009 workshop on Planning and Learning.

Reviewer for

• Artificial Intelligence
• Journal of Artificial Intelligence Research (JAIR)
• Journal of Scheduling (JOSH)
• AI Communications
• International Conference on Automated Planning and Scheduling (ICAPS) 2007-present; also, Doctoral Consortium mentor in 2009.

Invited Presentations

Applying AI Planning to Substation Voltage Control - Presentation to John Beddington CMG FRS (Chief Scientific Advisor to the UK Government), Royal Society of Edinburgh, February 2009.

Just for fun...

I have recently completed a module on Biological Psychology with the Open University, approaching the material from the viewpoint of how the subject area maps onto areas of AI.

When not working on AI, I have a strong interest in music. I enjoy playing the piano (completed a performance DipABRSM in 2004), and play the harpsichord in the University of Strathclyde Baroque Ensemble.