Ground Autonomy Evolution - Autonomous and Autonomic Systems - page 67

The following discusses the structure, architecture, behaviors, communi-
cation, and collaboration required to coordinate the activities of the agents in
supporting unattended mission operations in a satellite control center.
4.1.1 A Basic Agent Model in AFLOAT
An agent is a computer-based autonomous process that is capable of goal-
directed activities [ 17 , 45 ] and can accomplish tasks delegated to it with min-
imal reliance on human intervention. As it is goal-directed, it can allow the
user to specify simply what he/she wants [ 38 ], leaving to the agents how and
where to get the information or services. Each agent is also able to participate
in cooperative work as an associate with humans or as part of a community
of cooperating agents.
The agent architecture used in AFLOAT provided appropriate structural
elements and behavior to support basic requirements for adaptive reasoning
[ 172 ]. An agent is adaptive to the extent that it can respond to short-term and
long-term changes in its operational environment, deal with unexpected events
and reason opportunistically, maintain a focus of attention among multiple
goals, and select and perform appropriate actions.
Structural Elements of an AFLOAT Agent
Each agent in AFLOAT had three structural components (Fig. 4.1 ) :
1. An inter-agent communication interface
2. A monitor
3. A knowledge base
The inter-agent communication interface was responsible for validating
the inter-agent semistructured language format, sending outgoing messages,
receiving incoming messages, and broadcasting messages to other agents. The
monitor was responsible for monitoring interactions between agents, incom-
ing and outgoing messages, and the state of the agent, and maintaining a
history of the agent's actions. An agent's history of past-actions supports the
agent's learning from experience when presented with new tasks. Each agent's
knowledge base consisted of three elements:
1. A strategist, or a decision-theoretic planner
2. Problem-specific context descriptor
3. A set of procedures or rules for domain-dependent actions
The strategist was responsible for planning and scheduling the actions that
an agent must perform to achieve its goal. The problem context descriptor
defined specific attributes of each request to ensure that each agent's atten-
tion was focused on the problem at hand. Problem solutions were modeled
as domain-dependent procedures. The internal models maintained functions
(such as managing access to the skills of each agent or maintaining its message
buffer) that were private to each agent and not accessible to external agents.