The DPOL Project is conducting fundamental and applied research into operations/project level planning, with an emphasis on integrating methods from machine learning and optimisation. This four way collaboration between DSTO, University of South Australia, and University of Adelaide has been particularly successful at dealing with uncertainty in planning.
Operations planning involves larges groups of people choosing and co-ordinating tasks to produce a smoothly orchestrated operation. Automatically developing robust plans with hundreds of tasks is hard. Planning becomes even harder when trying to take the uncertainty of the world into account. NICTA (National ICT Australia), the Defence Science Technology Organisation (DSTO), the University of Adelaide (UofA), and the University of South Australia (UniSA) are developing theoretical frameworks, algorithms and tools that formalise, abstract, and solve such planning problems. The project draws from, and contributes to, the fields of machine learning, optimisation, control, and classical planning.
Theoretical contributions use tools such as category theory to establish links between these fields, describing equivalence classes between different representations of the problem, for example, Petri-Nets and Markov Decision Processes.
Algorithmic work has explored a wide cross section of methods previously used in planning, including Markov decision processes, SAT based planning, planning graphs and search, Petri-net unfolding, predicate logic, temporal logic, and optimisation methods; all resulting in a diverse selection of publications.
Concrete project goals include replacing commonly used tools, such as Microsoft Project, with software that automatically plans and schedules a set of tasks drawn from a database. Contributions so far have developed four separate planning servers to support existing military planning tools developed by DSTO. More tools are on the way, which will provide solutions to a broader range of problems than just planning under uncertainty.
Methods emerging from the project have application to the broader planning community, operations researchers, control theorists, and the day-to-day project managers who would like to know how a 50% chance of rain could affect their project budget. DPOLP work is also becoming concerned with the presentation of planning information, including theoretical work in how to measure the similarity of plans, and how to present qualitatively different plans to the user from a spectrum of valid plans. Beyond traditional operations and project planning, DPOLP tools for the analysis of uncertainty contribute to the decision support and business planning domains.
Key Achievements
- world wide leading implementations of probabilistic temporal planners
- 1st and 3rd in International Probabilistic Planning competition 2006
- Delivered three planning servers to DSTO
- Demonstrated in military training exercises
- 46 international publications and counting
- Interactions across five institutions
- NICTA now one of the world’s top planning groups
The following people are involved in the DPOL Project:
- Jussi Rintanen (Project leader, NICTA)
- Patrick Haslum
- Alfredo Gabaldon
- Sylvie Thiebaux
- Lewis Warren (Project leader, DSTO)
- Prof. Lang White (Project leader, University of Adelaide)
- Prof. Jonathan Billington (Project leader, University of South Australia).
Please have a look at the 2006 annual report