Simulated Tactical Air Flight using Soar
Tactical Air Flight using Soar
TacAir-Soar
(Jones, et al, 1993)
is a Soar system that models the
behavior of pilots in beyond-visual-range, tactical air combat.
The domain is
simulated,
although humans pilots may interact with
the Soar agents through simulators as well.
The domain represents an example of one of Soar's
methodological
assumptions: the
architecture is being pushed to the extreme by
forcing the integration of many
capabilities that have been fully or
partially demonstrated in Soar but never combined in a single system.
For example, agents are currently anticipated with the following
capabilities (TacAir-Soar is an on-going project and many of these
capabilities are currently being implemented and integrated):
- Reasoning
about High-level Cockpit Flight (e.g., turning)
- Spatial and Temporal Reasoning
- Planning
- Situational Awareness (Rapid Situation Interpretation and Assessment)
- Natural Language Communication
- Explanation/Justification of Actions Taken (Debrief)
- Dynamic Generation of
goal-based behavior
- Plan Recognition/Event Tracking
- Learning: Speedup, Induction, New Tactics
(from generalization), Achievement
of Multiple Goals
- Coordination and Cooperation with Other Agents
The last capability, cooperation and coordination, is relatively unique
in cognitive architectures, although not in the broader scope
artificial intelligence.
Cooperation occurs in this domain among a lead and his wingman.
In general, these two planes
fly and execute missions together, using radio communication to coordinate
their activity. Coordination is also driven from air (or ground) control,
which informs agents about enemy planes beyond their local radar range and
may order them to interdict specific agents. Thus, agents in this domain
may act according to some pre-specified mission, may recognize and
act autonomously to threats, and may be
tasked by other agents to accomplish
some specific action (such as interdiction or rendezvous).
Finally, integration of these capabilities results in many research
issues.
One of these is the problem of interacting goals. Soar is able
to consider
multiple, simultaneous
goals through the creation of the
context stack. However, this data
structure is hierarchical. Integration of multiple behaviors has shown
that there are times when goals could be better represented as constraints
and that separate goals may drive conflicting behaviors. Decisions about
how to represent non-hierarchical goals may drive changes in the
architecture.
Return to the top of this architecture.
Soar - Capabilities - Simulated Tactical Air Flight