jason-users Mailing List for jason

Hi John,

Sorry for the delay in authorising your message (the list is moderated
to avoid spam).

Now I understand better your problem.

> I have only one agent and but I fudge more agents by attaching an ant
> ID to every action, goal and event. In the environment I have a
> representation of multiple ants. So, when I get a move action, I know
> which ant representation to move. This is messy, but it works for now.
> eg.
>  +pos(ID,X,Y) : not carryingFood(ID)
>    <- randomMove(ID,X,Y).

Yes, this is a good solution (given the problems with percepts - more
about this below). You could also get the ants having a belief about
their own ID (e.g.: "self(ant1)."), and in the plans you make sure the
ID in question is that of self, so that they only act on what's relevant
to themselves.

Of course this isn't ideal as you rightly point out. The idea is to
have "customised perception" rather than "global". Note that there is
currently a mechanism to customise perceptions, which is the "perceive()"
method that can be redefined if you specify a customised agent
architecture. The environment in itself is "global", which makes
sense. The way we initially thought things, which is currenlty
implemented, is that all such customization should be done when the
agent gets perceptions. So the perceive() method gets everything that's
true about the environment, but in a customised architecture you remove
all the facts that are not perceptible to one agent particularly.

Problem is that this is inefficient, as in some applications, if you
have a large environment description, lots of agents, and limitted agent
range of perception, you get lots of facts being transmitted in the
network without need. That's why in 0.4 we already added the possibility
to customise this together with the enviorment. So when you define the
enviroment, you can redefine getPercepts() so that it check which agent
is actually trying to observe the environment and only let that agent
access (via getPercepts()) what is really specifically observable to it.

Version 0.4 is almost ready, but we didn't update the manual yet. We
hope to make it available next week, but we can't be sure as we are all
to busy with other things.

> I'm sure I can figure something out. I'm mainly just curious as to why
> percepts are global. Surely the agent doesn't precieve the whole
> environment?
Yes, you are right. As from 0.4 there will be two levels of
customisation: 
- getPercepts only gets the fact of the env. that are relevant to the
  agent (customisation of this is done in programming the environment).
- when the agent's own percepts get to it, you can still redefine
  perceive() to add, e.g., noise (something is accessible to the agent,
  but its perception is faulty).

> Why is BDI not a good thing to use for implementing an ant colony?
Because one tends to think of an ant as somthing not very cognitive
(i.e., they don't have beliefs and plans, they just react to
perception). But AgentSpeak is OK for this too, it's just that there are
lots of packages for reactive agents around, and they are a lot easier
to use (or maybe not!).

> I'm just learning all this, so please forgive any ignorances.
You learn very quickly, your comments were quite right!

> I'm enjoying learning Jason/AgentSpeak(L). Thanks for making it
> available.
Thanks for the feedback, this is much appreciated!


All the best,

Rafael

Hi Rafael,

Thanks for your quick response.

I have subclassed Environment and represent food and ants in it. I have
actions for moving and picking up food. Currently the agent just moves
randomly and until it finds food and then it makes it way back to base. It
all works fine. The problem is having multiple agents.

I have only one agent and but I fudge more agents by attaching an ant ID to
every action, goal and event. In the environment I have a representation of
multiple ants. So, when I get a move action, I know which ant representation
to move. This is messy, but it works for now.
eg.
+pos(ID,X,Y) : not carryingFood(ID)
  <- randomMove(ID,X,Y).

Perhaps I could use beliefs instead? If the ant runs into food in the move
action, I could add a food belief and send a beliefUpdated event.

The perceive function is still global to all agents, so I would still have
to figure out some way of transmitting this percept to a particular instance
of the agent.

I'm sure I can figure something out. I'm mainly just curious as to why
percepts are global. Surely the agent doesn't precieve the whole
environment?

Why is BDI not a good thing to use for implementing an ant colony?

When is 0.4 coming out?

I'm just learning all this, so please forgive any ignorances.

I'm enjoying learning Jason/AgentSpeak(L). Thanks for making it available.

John

----- Original Message ----- 
From: "Rafael H Bordini" <R.Bordini@...>
To: "John Sietsma" <jsietsma@...>
Cc: <jas...@li...>
Sent: Thursday, April 15, 2004 2:43 AM
Subject: Re: [Jason-users] Global perceptions


> Dear John,
>
> I'm not sure BDI is the best thing for you to implement an ant colony,
> probably not. But if you wanna try Jason, you'd have to get your
> "design" right. One possible (natural) way of seeing this model is that
> pieces of food are objects, so they need a representation in the
> environment. The things that are true of the environment can be accessed
> by the agents when they do "perception". So an agent may believe it's
> close to food if that information is present in the percepts it gets
> from the environment. Then, moving towards that piece of food is a
> possible action that the ant agent may want to execute (thus your
> enviroment must be prepared to be changed when the agent does execute
> that action; e.g. the position of the ant will change in the physical
> environment). You may also want to check either whether to actually let
> an agent have access to a particular piece of information (e.g., the ant
> is too far to sense the food). For this you can customise the perceive()
> function in the agent architecture or you can wait for version 0.4 where
> we'll have yet another mechanism for this (by which the information is
> removed before the agent gets the percepts, as a result of sensing the
> environment).
>
> Cheers,
>
> Rafael

Dear John,

I'm not sure BDI is the best thing for you to implement an ant colony,
probably not. But if you wanna try Jason, you'd have to get your
"design" right. One possible (natural) way of seeing this model is that
pieces of food are objects, so they need a representation in the
environment. The things that are true of the environment can be accessed
by the agents when they do "perception". So an agent may believe it's
close to food if that information is present in the percepts it gets
from the environment. Then, moving towards that piece of food is a
possible action that the ant agent may want to execute (thus your
enviroment must be prepared to be changed when the agent does execute
that action; e.g. the position of the ant will change in the physical
environment). You may also want to check either whether to actually let
an agent have access to a particular piece of information (e.g., the ant
is too far to sense the food). For this you can customise the perceive()
function in the agent architecture or you can wait for version 0.4 where
we'll have yet another mechanism for this (by which the information is
removed before the agent gets the percepts, as a result of sensing the
environment).

Cheers,

Rafael

Hi,

I have been implementing a simple agent systems for a university =
assignment. I decided to experiment with a any colony, as I thought this =
would be a nice and simple way to get to know Jason and BDI.

My question is, why are percepts a property of the environment? If an =
ant stumbles on some food I just want that particular ant to know about =
it. In executeAction(), when I move an ant(agent) and it comes across =
food I add a percept to the environment. Should I be adding to the =
agents beliefs instead?

Thanks for your help.
John

I forgot to say that you also need to import saci launcher classes:

	import saci.launcher.*;


Jomi

Jomi Fred Hubner wrote:
> Dalton,
>=20
> you can use the saci launcher to control the agent's life. The followin=
g=20
> example shows how the environment class could get more information abou=
t=20
> the running agents and how to kill some:
>=20
>=20
> class testEnv implements Environment {
>     .......
>=20
>     int count =3D 0;
>=20
>     void killAg1() {
>         count++;
>         if (count =3D=3D 10) {
>             try {
>                 Launcher saciLauncher =3D LauncherD.getLauncher();
>                 if (saciLauncher !=3D null) {
>                     Iterator iAgents =3D=20
> saciLauncher.getAllAgentsID().iterator();
>                     while (iAgents.hasNext()) {
>                         AgentId aid =3D (AgentId)iAgents.next();
>                         if (aid.getName().startsWith("ag1")) {
>                                 System.out.println("*** killing ag1 id =
=3D=20
> "+aid);
>                                 saciLauncher.killAg(aid);
>                         }
>                     }
>                 }
>             } catch (Exception e) {
>                 e.printStackTrace();
>             }
>         }
>     }
> }
>=20
>=20
>=20
>=20
>=20


--=20
WYSIWYG: what you see is all that you get
LaTeX: what you get is more than you see
--=20
Jomi Fred H=FCbner             Universidade Regional de Blumenau    - FUR=
B
Professor                    Departamento de Sistemas e Computa=E7=E3o - =
DSC
                              Campus IV, rua Braz Wanka,  238,  Vila Nova=

mailto:jo...@in...      89035-160                      Blumenau, SC
http://www.inf.furb.br/~jomi Tel: (47) 321-7810       Fax: (47) 321-7802

Dalton,

you can use the saci launcher to control the agent's life. The following 
example shows how the environment class could get more information about 
the running agents and how to kill some:


class testEnv implements Environment {
     .......

     int count = 0;

     void killAg1() {
         count++;
         if (count == 10) {
             try {
                 Launcher saciLauncher = LauncherD.getLauncher();
                 if (saciLauncher != null) {
                     Iterator iAgents = 
saciLauncher.getAllAgentsID().iterator();
                     while (iAgents.hasNext()) {
                         AgentId aid = (AgentId)iAgents.next();
                         if (aid.getName().startsWith("ag1")) {
                                 System.out.println("*** killing ag1 id 
= "+aid);
                                 saciLauncher.killAg(aid);
                         }
                     }
                 }
             } catch (Exception e) {
                 e.printStackTrace();
             }
         }
     }
}

Hi,

I forgot to add in the previous message that if you don't use any
annotation in the initial beliefs that are used in the AgentSpeak
source code, those are automatically interpreted as [self], thus
having the effect you wanted in that example.

Cheers,

Rafael

Hi Dalton,

> Which is the function of self in the beliefs.  For example:
> myself(ag3) [ self ].  Exists some difference in declaring an initial
> belief or in execution time (in executeAction).

Yes, there is an important difference between internal and external
beliefs. An internal belief persis until the agent removes that belief
explicitly (i.e., with -b(t) in the body of a plan). If a belief arises
as perception of the environment, it will cease being a belief as soon
as the environment sends a list of perception that does not include that
predicate and the agent does (the default) belief revision. So, if your
environment told the agent it's name, it would have to keep saying it
all the time, or the agent would "forget" it. That's why it's important
that the environment only sends beliefs about things in the environment
itself: the name of the agent, if necessary for the agent reasoning,
must be a belief of the agent itself, not one acquired from perception
of the environment.

Cheers,

Rafael

Hi,

Which is the function of self in the beliefs.
 For example: myself(ag3) [ self ].
Exists some difference in declaring an initial belief or in execution
time (in executeAction).

thanks, bye.

Oi,
   num exemplo definido no Jason eu tenho dois agentes e alguns "clones"
destes agentes.
   Depois de um certo tempo de processamento (dentro do executeAction)
eu gostaria de matar uma inst=C3=A2ncia (um clone) deste agente...

Algu=C3=A9m poderia me ajudar..

obrigado, bye

A new version of Jason is available at

https://sourceforge.net/project/showfiles.php?group_id=98417

This version is more userfrendly and supports windows (it has .bat scripts)

The program is started by the Jason/bin/jason.bat script. To run an 
example, select the menu option "Project|Open", select the .mas2j file, 
select the menu "Project|run".

Regards,

Jomi