Environmental Engineering Reference
In-Depth Information
can receive real data from the other levels. The following subsections discuss
each of these parts in more detail.
9.6.1 Agent Development
The lower part of Fig.
9.5
represents the agent-development level, where de-
velopers write the code, modify existing agents, or use an automated agent-
development system that searches for previously developed agents that per-
form a needed task and updates them based on new requirements input from
the developer.
Domain-specialist agents are available to assist in the development of new
agents by interacting with the agents being developed, giving a new agent
other agents to interact with and then testing for proper functionality. In ad-
dition, data may be received from operational agents in the ground control
system and spacecraft for additional testing purposes. Agents in the opera-
tional mode would know that messages from agents under development are not
operational by virtue of a marking of the messages by the messaging service
that passes messages to the operational agents. As agents are developed, they
are added to the community of domain experts and provide the developers
with additional example agents to modify and test against.
The agent-development level also represents an agent incubator. After an
agent is developed, there is an incubation period during which agents are
tested in a background or shadow mode. When confidence in the agent's be-
havior is attained, it is moved into an online community doing real work in
its domain. It is at this level that the credentials of the agent come into play.
These credentials attest to the development methodology and the verifica-
tion and validation procedures that would directly ensure the agent's correct
behaviors.
9.6.2 Ground-Based Autonomy
The middle section of Fig.
9.5
comprises two parts. The right side represents
the ground control system, to which some of the agents may migrate after
development. In this part, the agent can run in a shadow or background mode
where its activities can be observed before it is put into full operation. This
allows the users to gain confidence in the agent's autonomy before committing
to or deploying it. If a problem is found with the operation of the agent or its
operation is not as envisioned or required by the end user, or if enhancements
are desired, the agent (or copy of the agent) can be migrated back down to
the development area for further modifications or testing. Once modifications
are made, it can then be sent back up to the ground operations level, and the
process is repeated.
The right side of the middle section also contains an area where agents can
be cloned to support parallel processing or fault tolerance: identical agents can
be run on multiple (even geographically distributed) platforms.
