Plan representations have been developed over several decades of AI planning research [2]. They can support a rich model of processes, tasks, plans, resources and agents. These representations can be used for purposes other than plan generation. Some key concepts include hierarchical plan representations, rich activity and resource models, the capturing of the intentions behind plan steps, and languages or ontologies in which to express the activity and process models.
Knowledge rich plan representations such as those in the Edinburgh O-Plan [4] and O-Plan2 [8] planners have been used successfully in a number of projects.
The PLANIT work [5] is a prototype system produced during the UK Alvey Programme in which rich plan representations were used without plans being actually generated. In PLANIT, flexible plan representations provided integration across an enterprise involving project management (interfaced to the ARTEMIS system), process planning (interfaced to a Jaguar Cars' process planner) and job shop scheduling (interfaced to the UK Atomic Energy Authority's WASP scheduler). PLANIT could help the user to browse on a plan, monitor its execution and make single step modifications to it as necessary, taking into account knowledge of resources, agent capabilities, how the original plan was constructed and what the aims of the plan were.
OPTIMUM-AIV [1] is a more recent example of the use of flexible plan representations in a project management domain alongside ARTEMIS project support tools. OPTIMUM-AIV is a flexible planning and re-planning system for spacecraft assembly, integration and verification at the European Space Agency.
These two systems explicitly represent the causal structure of a plan, to hold the dependencies between the preconditions and effects of activities involved in the plan -- therefore showing the rationale or intentions behind the plan. Dependencies of the same kind are useful in all aspects of plan generation, execution monitoring and plan repair.