Coalition, from Latin coalescere (co-, together + alescere, to
grow) is a type of organisation where joint members work together
to solve mutual goals. The principal feature of a coalition is the
existence of a global goal, which motivates the activities of all
coalition members. However, normally such members are not directly
involved in the resolution of this goal, but in sub-tasks
associated with it. Typical features of a task scenario that requires the deployment
of a coalition are:
- The nature of the problem demands a set of different abilities;
- Participants need to collaborate because they have limits in abilities and knowledge;
- The task requires a structure of command and control that coordinates
the activities of participants. Coordination is also essential to avoid conflict
and improve the use of time and resources;
- There is the problem of integrating heterogeneous systems belonging to multiple
organisations with distinct doctrines and operational rules.
Examples of Coalition Scenarios
Urban disaster relief domains require the union of agents that support
activities related to search and rescue. This coalition instance (left-hand side) is based
on the RoboCup Rescue Simulator . Agents are represented by fire brigades,
ambulances, police forces, fire stations, ambulance centres, police offices and a search
and rescue command centre. The scenario could represents a real part of Kobe City (Japan).
The next figure (right-hand side) illustrates an example of a military domain where multinational forces integrate abilities during operations of peacekeepers, such as in the Iraq scenario.
Agents are represented by the head country's Department of Defence, local operational
agents and several military units. Multinational coalitions are a typical example
of coalitions that require the development of shared representations reflecting different
cultures, doctrines and languages. Groups such as the Multinational Planning Augmentation
Team (MPAT) work in this direction, trying to develop and become familiar with Standard Operating Procedures (SOP), which represent well-founded and tried plans to be applied in specific situations.
The last figure illustrates a futuristic scenario where robots and astronauts collaboratively
work in interplanetary missions, such as a Mars Mission. Agents are represented
by the Earth Mission Control Centre, Mars Habitats and units of astronauts
and robots. A real example in this context is the Aurora Exploration Programme , a long-term effort of the European Space Agency (ESA) that aims to send a team like that to Moon in 2024 and to Mars in 2033.
This thesis, in particular, is concerned with the role of intelligent planning in coalition
support systems. Planning can bring several advantages to coalition operations such
as prediction of failures, resource allocation, conict identication and so on. The
planning process in coalitions is naturally distributed because each coalition member
is a decision-maker.
We are delivering planning mechanisms to users via assistant agents. In this approach
each participant has an agent that supports his/her tasks, providing for example, planning
information and options to carry out activities. This approach is powerful because
while users have the ability to take decisions based on their past-experience (case-base
reasoning), agents are able to generate and compare a significant number of options,
showing both positive and negative points of such options.
Agents that support coalition members at different levels of
decision-making must be customised so that they are able to support the planning activities carried out at each level. From a practical perspective, this customisation means that agents must provide activity handlers (e.g. pathfinder, load balancing, etc.) to support a specific set of activities (see figure below). Such handlers work on a planning representation, which expresses, for example, notions of environment, time, resources, priority, activity, state, etc.
However, when agents are performing as part of a coalition, the complexity of the
planning process increases due to the number of requirements that must be considered.
Requirements associated with collaboration, distributed planning, coordination
and user interaction are interrelated and the design of such one can inuence others.
Thus the implementation of individual solutions for each set of requirements is not a
good practise. Part II of this thesis analyses in details such requirements and their role
inside the planning representation.
The principal claim of this work is that we can integrate distinct groups of requirements
(multiagent planning, collaboration and human interaction), associated with the
development of hierarchical coalition support agents, via a unied framework provided
by a constraint-based ontology and related functions. We argue and demonstrate
that such framework brings several advantages for the agents' development, such as: well-known environment to represent and build plans; transparent way to incorporate
collaborative concepts, which complement the planning abilities; opportunities for the
development of more advanced human-agent mechanisms, and; support to an easy
customisation of activities handlers. In brief, we have the following goals:
- Formalisation of hierarchical coalitions (members, relations and rules among
them) so that we can use the structural features of this kind of organisation for
the development of command and control mechanisms (coordination);
- Investigation and categorisation of requirements that have inuences on the development
of models and processes used by agents operating in hierarchical structures;
- Specification of a unified representation of planning and collaboration that enables
an easy customisation of activity handlers and an appropriate basis for the
incorporation of requirements associated with human-agent interaction;
- Development of practical applications that demonstrate the real advantages of
this approach and also stress its generality.
The use of hierarchies facilitates the deployment of coordination mechanisms because
such mechanisms can exploit the hierarchical organisational structure. This is because such an organisation implicitly denes the agents responsibilities, capabilities, connectivity and control flow. In addition hierarchies also have the following advantages:
Hierarchical organisation can be classified into three levels of decision-making: strategic, operational and tactical. This hierarchical arrangement is a common practise in military
models of command and control and consistent with knowledge engineering work. Furthermore, this classification also highlights the need of customising planning processes at each of these levels.
- They are compatible with the divide-and-conquer idea. The process of splitting
a problem into smaller subproblems is repeated at each level;
- Hierarchical levels may deal with different granularities of knowledge so that
each level does not need to have all the details about the problem;
- It is possible to enclose problems to be dealt with by local subteams, instead of
spreading it over the coalition.
A formal specification and definition for the hierarchical coalitions that we are work with can be seen here.
- The strategic level accounts for developing plans in a high-level of abstraction, or
"what-to-do" plans. In other words, the level specifies what must be done, but it does
not give details about how something must be done. In this way, the principal tasks
are related to analysis, directions and comparison of courses of actions.
- The operational level accounts for refining the plans produced at the strategic level,
mainly providing the logistical resources for them via processes of resource scheduling
and load balancing. Thus, knowledge about the operation environment is more detailed
and limited coalition groups will be affected by the decisions.
- The tactical level is where the execution of operations actually takes place. For this
reason the degree of knowledge of tactical components is very specialised on the domain that they are operating, and their decisions are generally taken on sets of atomic
activities. As the components are performing inside a dynamic and unpredicted environment,
their reactive capabilities and speed of response are very important so that
the use of pre-dened procedures could be useful alternative. The result of this level
are atomic activities or set of atomic activities that are commonly executed by the own
Centre for Intelligent Systems and their Applications
School of Informatics, The University of Edinburgh
Page maintained by
Last updated: Tue Oct 18 23:59:50 2005