Screen shot (23k)
Tachyon is a constraint-based system for temporal reasoning,
capable of reasoning about both qualitative and quantitative aspects of time.
Tachyon is being developed at GE CR&D with partial funding from ARPA and
US Air Force Rome Laboratory as part of their planning and scheduling
initiative.
For a summary of some of our recent work, take a look at the slides from our recent talk at Rome Laboratory.
OPERATING ENVIRONMENT
OPERATING SYSTEM:
Sun OS 4.1.3
X-WINDOWS:
X11R4 or later
HARDWARE:
SUN SPARC series
MEMORY REQ:
Minimum of 32MB RAM, 1 GB Hard disk.
STATUS OF SYSTEM
Prototype, has been integrated with several other
prototype systems, available upon request.
AVAILABLE SYSTEM DOCUMENTATION/REFERENCES
User Guide, Tutorial, Object Description Document.
POINTS OF CONTACT
Jonathan Stillman
GE Corporate Research & Development Center
(518)387-5257
stillman@crd.ge.com
James Farley
GE Corporate Research & Development Center
(518)387-6513,
farley@crd.ge.com
Richard Arthur
GE Corporate Research & Development Center
(518)387-6823,
arthurr@crd.ge.com
DESCRIPTION
Tachyon provides powerful temporal reasoning capabilities that can be
applied effectively to a number of critical problems. We have concentrated on
Tachyon's use in refinement and feasibility tracking of the temporal aspects
of courses of action (projecting critical steps in a deployment plan to insure
proper interaction between them), and on preserving temporal consistency
during force package structuring and deployment (used in coordination with
ForMAT and/or TPEdit, a TPFDD editor). We have also applied Tachyon to plan
recognition tasks, where it was used to validate temporal sequencing of events
as an aid in formulating plan hypotheses, to plan generation and monitoring,
to scheduling for plastics and power systems manufacturing, and to retrieval
and situation refinement in a prototype spatio- temporal data management
system. Tachyon's constraint propagation capabilities were used there together
with partial information about interrelated events to provide temporal
refinement for tasking support. We have also developed a prototype integrating
Tachyon with CAFE, GE's case-based reasoning system for constraint- directed
force expansion.
Some of Tachyon's key features are summarized below:
- ability to cope with incomplete information regarding all aspects of events, e.g.,
start-times, finish times, duration, and inter-event constraints
- ability to express both qualitative and quantitative constraints, e.g., X is before
or meets Y, and X is before Y by 12 to 15 hours
- efficient storage and retrieval of complex temporal data items, supporting
queries such as "Which events could possibly be occurring between 8:00 Tuesday and
14:00 Wednesday?"
- scalability, providing efficient propagation of constraints even in graphs with
tens of thousands of nodes and edges
- an intuitive, easy to use graphical user interface
- versatility of both data model and software (Tachyon's model is quite powerful
and versatile, and the software can be used stand-alone or as an embedded
application.)
- support for situation refinement and other exploitation tasks.
Tachyon is provided with a graphical user interface (GUI) to facilitate
interaction on a dynamic level. One can use Tachyon to test consistency of a
network, fine tune it, or test it in ``What if...'' scenarios as desired, then
save the network. The interface itself is a CAD-like direct-manipulation
editor for the graphical representation of the underlying network. We have
also isolated the interface itself to provide a generic graph editor
library. Tachyon is available in a ``batch mode'' form, in which interaction
is performed via text files (that may be created using the GUI or by another
process).
As appropriate for the application. Tachyon takes as input a specification of
a temporal constraint network that specifies events (tasks), windows of
feasibility for events, and inter -event constraints. Currently, input is
provided either via the GUI, files, or RPC communication from another process.
Tachyon produces a refined version of the temporal constraints that expresses
the maximal feasible times of occurrence of the events given the input
constraints.
