From: Ranjit Nair (nair@usc.edu)
Date: Thu 17 Oct 2002 - 05:42:20 GMT
I tend to agree with Mazda on the issue of the ACL. It is not for the
simulator to specify the format or content of messages. In order for the
simulation to be realistic, the agents should be free to decide on how
or what they communicate. Hence I am against the idea of an explicit ACL.
It should be up to the agents to decide on whatever ACL to use.
This said, the kinds of restrictions you propose are very necessary too. I
would thus put restrictions on:
1. The number of messages sent (already exists)
2. The number of messages received (already exists)
3. The number of bytes communicated. (new proposal)
This could be a restriction on the
size of an individual message or the total number of bytes sent in a
single cycle. In most human languages we talk at about 5-6.5 syllables/sec.
This translates into roughly 300-400 syllables/minutes. The average word
in English is 1.4 syllables long and roughly 4-5 bytes (including spaces).
Therefore, if we were considering human speech in English we would be able
to speak about 1100 bytes if we were speaking continuously. That is no
time for acting or listening.
Let us assume for a moment that we are allowed to hear and send 4 messages
of 80 bytes each in every time step. Since speaking rates= listening rate,
this would mean about that for 8 messages (4 heard and 4 spoken), it would
take the agent about 35 secs leaving about 25 secs to act. I think this is
a fair assumption.
Hence my proposal is that each message that an agent send should not be
more than 80 bytes long. Such restrictions on the amount of information shared prevents the
distributed problem degenerating into a centralized single controller
problem.
I think that centers (fire station, ambulance
center and police office) should be allowed to send and receive more
messages than other agents because we can assume that they have better
hardware and more manpower. I agree with Takeshi's proposal on the number
of messages for centers.
Regards,
Ranjit
> Dear All,
>
> > My main question is why there is a need of definition of an explicit ACL for
> > agent conversation? If you think there is need, I suggest KQML.
>
> The unlimited communication makes the simulation not a multi-agent
> problem but a single-controller problem. The unlimitation enables
> agents to share information and their decision. Multi-agent systems are
> more appropriate for RoboCup-Rescue, originated by RoboCup-Soccer.
>
> On the other hand, to simulate or to exceed human intelligence is the
> great goal of AI. The main purpose of the competion is to provide an
> evaluation of AI methods, so it is a good choice to try to simulate
> human's restriction.
>
> Regards,
> Tetsuhiko Koto
> koto@takopen.cs.uec.ac.jp
>
>
This archive was generated by hypermail 2.1.4 : Thu 17 Oct 2002 - 06:12:36 GMT