I'm planning on sending the note below tomorrow to Doyle, etc. Please
know if you have any corrections, additions, etc. Hope I haven't misrepresnted
This message is intended to briefly describe the TIE that I promised
to write up after the meeting at ISX last week. If you recall, this TIE
was tentatively called something like "Find Civilians, Get Them to Embassy",
for lack of a better name :-) The scenario involves finding
a group of protesters who are staying at various locations in Kuwait City,
and bringing them to a temporary haven in Kuwait city -- the Embassy.
From there, they will be shipped off to the States. (The TIE being
written up by Miland Tambe takes off at this point.)
The TIE currently is planned to involve the following participants:
ISI - Ariadne (Steve Minton, Craig Knoblock)
SRI - OAA & Multimodal Map System? (Adam Cheyer, David Martin)
OBJS (OSCI) - WebTrader (Paul Pazandak, Tom Bannon)
CMU - Simulator & Prodigy? (Manuela Veloso, Tucker Balch)
Since the meeting last week, I've communicated with each of the groups
listed above reach a basic agreement on the scenario and begin
considering the technical issues. Some of the boxes above have still
question marks next to them, as described below.
Basic NEO Scenario
A group of American civilians, presumably protesters, need to be found
and evacuated from Kuwait. First, we need to identify where the
civilians are staying in Kuwat. Ariadne and Webtrader will work
together to accomplish this task. Specifically, Webtrader will act as
an information broker for Ariadne by locating some of the required
information sources. Ariadne will handle the actual information
retrieval. To accomplish the task, an embassy information source
(e.g., a database containing visa information) will be accessed to
determine where each of the civilians is staying (e.g. hotel, private
residence, etc.). Then a second set of sources (probably on the web)
will be accessed to obtain the corresponding addresses and phone
numbers that can be used to contact the civilians. Finally, a
web-based geocoder will be used to convert these addresses to Lat/Long
The addresses will then be handed off to Prodigy, which will plan an
evacuation route (or routes) for picking the civilians up and bringing
them to the Embassy. (Note: Manuela may substitute another planner she
has developed.) The evacuation to the embassy will be simulated using
the CMU simulator. Unexpected roadblocks or other problems will be
encountered along the way, and the planner will need to dynamically
SRI's OAA will be used as the agent framework for this TIE. All of the
agents above will thus communicate using via OAA.
One point that came up, as we attempted to refine the scenario, is
that we will also need a human interface which will "control" the
demo and enable humans to conduct this operation in a sensible fashion.
For instance, in a realistic scenario, the military might try to
contact these civilians by phone before picking them up. In
particular, one would want to ascertain if the address information was
correct and determine if there was anyone left off the list, etc.
Thus, we are considering incorporating SRI's multi-model interface
that Adam Cheyer demonstrated. A human could use the interface to
request that the information agents find the current locations of the
civilians, and then call the planner to plan the evacuation route(s)
Unfortunately, I have not been able to talk to Adam directly (he has
been in Switzerland), but Karen Myers and David Martin at SRI felt
that the interface Adam demoed would be an appropriate choice. (Note
that it already runs in OAA.)
We could make the information agents' job more interesting by assuming
that we only know the name of the protest group ("Stanford Students
for the Conservation of Fossil Fuels" :-). Then we could take a roster
of possible members of this group, and intersect it with recent
visitors entering Kuwait to obtain the names of the civilians we need
to evacuate. Matching names involves recognizing matches that are not
identical (e.g. Steve Minton vs. S.N. Minton); this is something we
can do with Ariadne.
The CMU folks would like to show off the dynamic replanning
capabilities of their system. The simplistic way to do this is to have
the simulator "discover" roadblocks or other difficulties. A more
interesting possibility is for the human controller to add/delete
locations while the simulator is executing. This would make sense in a
scenario where humans take the addresses returned by the information
agents, and begin verifying the locations of the civilians by phone
while they simultaneously plan the route(s) and begin evacuating
people. Another possibility is to have Ariadne (or some other
information agent) monitor traffic conditions.
Note: When the addresses are handed off to Prodigy, Ariadne will
classify each address as a hotel, residence, etc, since this may be
useful for the planning & simulation. (Sorry, I'm now a bit fuzzy on
how this would be used, but Tucker had something in mind.)
Here is a short sample of technical issues. I'm sure there will be many
but these are a few that have come up in our preliminary discussions.
- As noted above, we need to determine exactly how the human interface
be used, and how it we will communicate requests to the information agents
and the planner.
- We need to determine the right way for Webtrader and Ariadne to communicate.
Currently the idea is that they will both be OAA agents it might make
sense to use a web-based communication channel.
- The CMU folks requested that addresses be given as X,Y coordinates
map, rather than lat/long. The transformation can be computed by Ariadne,
but we should consider what is the "right" way to accomplish this.
At this point, I think it would be helpful to have some assistence
refining the scenario to make sure it is realistic.
We need to decide on the precise information sources that will be identified
and used by WebTrader and Ariadne.
The suggestion was made that we should set up a regular (bi-weekly?)