Review of State of the Art in Planning and Scheduling as
Related to COORDINATORS Project: Part I
December 22, 2005
Summary of Program Concept
The COORDINATORS program emphasizes distributed intelligent, cooperative problem solv-
ing to a degree not previously demanded of the research community. Automated agents must
recognize important change, assess its ramiﬁcations on its local view of activity, request or
propagate information to other agents as necessary to reduce uncertainty and accommodate
change, and re-schedule activity or select contingencies to stave oF plan failure. Additionally,
the agents must manage their own limited resources (computational and informational), make
decisions within the allowed rule bounds and learn models of appropriate interaction with their
human users. All of these tasks must be accomplished within an intrinsically distributed envi-
ronment with interaction and knowledge circumscribed by formal organization such as chain of
command and need-to-know.
The BAA set out ﬁve primary technical areas: distributed activity coordination, context-
dependent coordination autonomy, machine learning, organizational reasoning and meta-cognition.
The BAA also identiﬁed four key hard research problems: distributed coordination over large
dynamic structures, coordination of multiple role units, learning appropriate decision making
autonomy with sparse data, adapting activity in real time in response to change and incorpo-
rating military decision policies in coordinated decision making.
In this document, we will review research in planning and scheduling that bears directly on
the hard research problems in COORDINATORS. The focus will be on identifying what most
taxes the current state of the art, what has been done previously to address those issues and
how the contractors’ research ﬁts in the ﬁeld as a whole.
Issues that Tax the State of the Art
The COORDINATORS vision goes well beyond the reach of current state of the art. Most of
the core capabilities required (e.g., reasoning about temporal changes in tasks, identifying al-
ternative resources, representing distributed task interdependencies, etc.) have been previously
developed to a limited extent. This is reﬂected in the proposals from the contractors, who are
working from an existing method base and integrating components and extending capabilities.
However, the capabilities have been studied either in a centralized framework, independent of
other capabilities, or in a small scale application with simpliﬁed characteristics (e.g., limited or