Approach
The O-Plan (the Open PLANning Architecture)
Project is exploring issues of coordinated command, planning and
control.
The objective of the O-Plan Project is to develop an architecture
within which different agents have command (task assignment), planning
and execution monitoring roles.
Each agent has a structure which separates the following components:
- the representation of the processing capabilities of an agent;
- the computational facilities available to perform these capabilities;
- the constraint managers and commonly used support routines which are
useful in the construction of command, planning and control systems;
- the decision making about what the agent should do next;
- the handling of communications between one agent and others.
The main contribution of the O-Plan research is to provide a complete vision
of a more modular and flexible planning and control system incorporating AI
methods.
We have demonstrated our approach on realistic problems related to
military logistics. Our approach will have an impact in the following ways:
- it will allow for improved connectivity and consistency between command,
planning and control;
- it will make plans open, inspectable, explainable and changeable;
- it will allow greater scope in Course of Action analysis and as such
provide greater plan reliability.
Within the agent-based O-Plan architecture we have created specific agents
to provide a domain-independent general planning and
control framework with the ability to embed detailed knowledge of the
domain. The system combines a number of techniques:
- A hierarchical planning system which can produce plans as
partial orders on actions similar to the Edinburgh Nonlin planner.
- An agenda-based control architecture in which each control cycle
can post pending tasks during plan generation.
These pending tasks are then picked up from the agenda and processed by
appropriate handlers (Knowledge Sources).
- The notion of a ``plan state'' which is the data structure containing
the constraints on emerging plan, the ``agenda'' of further requirements,
and the information used in building the plan.
- Constraint posting and least commitment on object variables.
- Temporal and resource constraint handling using incremental algorithms
which are sensitively applied only when constraints alter.
- O-Plan is derived from the earlier Nonlin planner from which it takes
and extends the ideas of Goal Structure, Question Answering (Modal Truth
Criterion) and typed conditions.
- We have extended Nonlin's style of task description language Task
Formalism (TF).
The Scenario
- A Task Assigner specifies a task that is to be performed
through some suitable interface.
- A Planner plans and (if requested) arranges
to execute the plan to perform the task specified.
- The Execution system seeks to carry out the detailed tasks
specified by the planner while working with a more detailed model of
the execution environment.
Page maintained by oplan@ed.ac.uk,
Last updated: Tue Nov 9 13:29:53 1999