adding a new KQML performative

[Vin Baines]

Hi All,

I'm using Jason agents in combination with the AgentScape platform,  
where I've got a number of agents running in agentscape acting as  
'sensors'. Jason agents are being succesfully told information by the  
agentscape agents and executing plans based upon this info, but I want  
to have the ability for Jason agents to query back to agentscape to  
verify/assert. I've been thinking about the easiest way to do this, as  
a bit of java code is going to be needed to deal with info flow  
(agentscape handles & the passing of info back to jason).

The easiest (and hopefully sensible) way I can think of doing this is  
to add a new KQML performative to extend ask - e.g. askAgentScape (for  
the time being - this assumes the jason agent knows to ask agentscape  
rather than just ask! - but it would prove the concept).

The implementation of jason inside agentscape currently doesn't handle  
any environments, so I can't just add new java code in there to add  
this extra functionality (e.g. as per room demo).

So, two questions really. Where should this be added in? I'm thinking  
extend jason with the functionality using agentscape classes, and then  
use the new jason.jar into agentscape so that the asl files can use  
something like askAgentScape(agName, value).

Second question - does this sound a sensible way of going about it?  
I'm keen to not go down a wrong hole for the 100th time..

Thanks

Vincent


------------------------------------------------------------------------------
Crystal Reports - New Free Runtime and 30 Day Trial
Check out the new simplified licensing option that enables unlimited
royalty-free distribution of the report engine for externally facing
server and web deployment.
http://p.sf.net/sfu/businessobjects
_______________________________________________
Jason-users mailing list
Jason-users@...
https://lists.sourceforge.net/lists/listinfo/jason-users

[Rafael H Bordini-2]

Hi Vin,

Jomi and I discussed this, and without knowing the details of the
interaction of Jason and AgentScape we can only try to guess some
possible alternatives:

1. Creating a new performative as you probably know is done completely
on the AgentSpeak side, but if we understand correctly you'll then need
to intercept this new performative and do something different with it in
the Java side. For this, what you need to do is to use a customised
AgArch class for your agents, and override the sendMsg method to use the
super-class method for normal messages and you handle the askAgentScape
in the way you want within your sendMsg method (with Java code that gets
the information from AgentScape and force the replies into the agent
mailbox, for example). There are some detail about architecture
customisation in the Jason book, and jason.sf.net/api will probably help
too.

2. If you have created a new Infrastructure which uses AgentScape to
create (and possibly distribute) agents (actually, I think someone in
the Netherlands did that, but I'm not sure), then you already have
methods for this and you'd need to change them to handle the new
performative.

3. Another possibility might be to use internal actions (which are
implemented in Java) to retrieve the info from the AgentScape side and,
e.g., add them in the agent's belief base or mailbox or however you want
the agent to have access to the results.

Sorry if this isn't helpful, but we don't know much about AgentScape and
the details of how you interfaced it with Jason. Feel free to ask
further questions and send more details if you think we can help.

Cheers,

Rafael and Jomi



Vin Baines wrote:

> Hi All,
>
> I'm using Jason agents in combination with the AgentScape platform,  
> where I've got a number of agents running in agentscape acting as  
> 'sensors'. Jason agents are being succesfully told information by the  
> agentscape agents and executing plans based upon this info, but I want  
> to have the ability for Jason agents to query back to agentscape to  
> verify/assert. I've been thinking about the easiest way to do this, as  
> a bit of java code is going to be needed to deal with info flow  
> (agentscape handles & the passing of info back to jason).
>
> The easiest (and hopefully sensible) way I can think of doing this is  
> to add a new KQML performative to extend ask - e.g. askAgentScape (for  
> the time being - this assumes the jason agent knows to ask agentscape  
> rather than just ask! - but it would prove the concept).
>
> The implementation of jason inside agentscape currently doesn't handle  
> any environments, so I can't just add new java code in there to add  
> this extra functionality (e.g. as per room demo).
>
> So, two questions really. Where should this be added in? I'm thinking  
> extend jason with the functionality using agentscape classes, and then  
> use the new jason.jar into agentscape so that the asl files can use  
> something like askAgentScape(agName, value).
>
> Second question - does this sound a sensible way of going about it?  
> I'm keen to not go down a wrong hole for the 100th time..
>
> Thanks
>
> Vincent
>
>
> ------------------------------------------------------------------------------
> Crystal Reports - New Free Runtime and 30 Day Trial
> Check out the new simplified licensing option that enables unlimited
> royalty-free distribution of the report engine for externally facing
> server and web deployment.
> http://p.sf.net/sfu/businessobjects
> _______________________________________________
> Jason-users mailing list
> Jason-users@...
> https://lists.sourceforge.net/lists/listinfo/jason-users

------------------------------------------------------------------------------
Crystal Reports - New Free Runtime and 30 Day Trial
Check out the new simplified licensing option that enables unlimited
royalty-free distribution of the report engine for externally facing
server and web deployment.
http://p.sf.net/sfu/businessobjects
_______________________________________________
Jason-users mailing list
Jason-users@...
https://lists.sourceforge.net/lists/listinfo/jason-users

[Vin Baines]

Hi,

Sorry for slow reply, I've been snowed under at work.

You're right, I'm building on the work you mentioned  
(http://www.iids.org/aigaion/indexempty.php?page=actionattachment&action=open&pub_id=261&location=thesis-Rogier.pdf-bfa451fd5967ae1e6b2d3d8f990dc368.pdf).

I'd thought it might be easier to use the existing ask performative,  
and use this in a send message, so that agents connected to a physical  
system in agentscape can report back data. e.g. send(sensor, ask,  
speed) so that any sensors currently existing in agentscape will  
report back values. Eventually I'm hoping to expand this out so that  
RMI type actions can be called on the physical sensor, using Jason to  
do the reasoning and then controlling sensors to improve, assert,  
disprove its belief base of what is going on in the real world.

So, a list is maintained in agentsapce of all agents, so i can use  
Map<AgentHandle, String> to =  
lookupPublished(e.getToHandle().toString()) to lift the 'to' component  
from the previous send message, which will then give me handle(s) to  
all the sensor agents in agentscape, and pass the message on.

The problem is, I can't get this message out to agentscape. I'm  
hitting the problem that Jason doesn't know that 'sensor' exists, and  
so does the sensible thing, reporting that the atom isn't valid.

Is there an easy way to override this, or will it break something in  
the jason architecture? Or a better approach?

Thanks for any suggestions!

Vin

Quoting Rafael H Bordini <r.bordini@...>:

> Hi Vin,
>
> Jomi and I discussed this, and without knowing the details of the
> interaction of Jason and AgentScape we can only try to guess some
> possible alternatives:
>
> 1. Creating a new performative as you probably know is done completely
> on the AgentSpeak side, but if we understand correctly you'll then need
> to intercept this new performative and do something different with it
> in the Java side. For this, what you need to do is to use a customised
> AgArch class for your agents, and override the sendMsg method to use
> the super-class method for normal messages and you handle the
> askAgentScape in the way you want within your sendMsg method (with Java
> code that gets the information from AgentScape and force the replies
> into the agent mailbox, for example). There are some detail about
> architecture customisation in the Jason book, and jason.sf.net/api will
> probably help too.
>
> 2. If you have created a new Infrastructure which uses AgentScape to
> create (and possibly distribute) agents (actually, I think someone in
> the Netherlands did that, but I'm not sure), then you already have
> methods for this and you'd need to change them to handle the new
> performative.
>
> 3. Another possibility might be to use internal actions (which are
> implemented in Java) to retrieve the info from the AgentScape side and,
> e.g., add them in the agent's belief base or mailbox or however you
> want the agent to have access to the results.
>
> Sorry if this isn't helpful, but we don't know much about AgentScape
> and the details of how you interfaced it with Jason. Feel free to ask
> further questions and send more details if you think we can help.
>
> Cheers,
>
> Rafael and Jomi
>
>
>
> Vin Baines wrote:
>> Hi All,
>>
>> I'm using Jason agents in combination with the AgentScape platform,  
>>   where I've got a number of agents running in agentscape acting as  
>>   'sensors'. Jason agents are being succesfully told information by  
>>  the  agentscape agents and executing plans based upon this info,  
>> but I want to have the ability for Jason agents to query back to  
>> agentscape to  verify/assert. I've been thinking about the easiest  
>> way to do this, as a bit of java code is going to be needed to deal  
>>  with info flow  (agentscape handles & the passing of info back to  
>> jason).
>>
>> The easiest (and hopefully sensible) way I can think of doing this  
>> is  to add a new KQML performative to extend ask - e.g.  
>> askAgentScape (for the time being - this assumes the jason agent  
>> knows to ask agentscape rather than just ask! - but it would prove  
>> the concept).
>>
>> The implementation of jason inside agentscape currently doesn't  
>> handle any environments, so I can't just add new java code in there  
>>  to add  this extra functionality (e.g. as per room demo).
>>
>> So, two questions really. Where should this be added in? I'm  
>> thinking  extend jason with the functionality using agentscape  
>> classes, and then use the new jason.jar into agentscape so that the  
>>  asl files can use  something like askAgentScape(agName, value).
>>
>> Second question - does this sound a sensible way of going about it?  
>>   I'm keen to not go down a wrong hole for the 100th time..
>>
>> Thanks
>>
>> Vincent
>>
>>
>> ------------------------------------------------------------------------------
>> Crystal Reports - New Free Runtime and 30 Day Trial
>> Check out the new simplified licensing option that enables unlimited
>> royalty-free distribution of the report engine for externally  
>> facing server and web deployment.
>> http://p.sf.net/sfu/businessobjects
>> _______________________________________________
>> Jason-users mailing list
>> Jason-users@...
>> https://lists.sourceforge.net/lists/listinfo/jason-users

------------------------------------------------------------------------------
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day
trial. Simplify your report design, integration and deployment - and focus on
what you do best, core application coding. Discover what's new with
Crystal Reports now.  http://p.sf.net/sfu/bobj-july
_______________________________________________
Jason-users mailing list
Jason-users@...
https://lists.sourceforge.net/lists/listinfo/jason-users

[Rafael H Bordini-2]

Ooops, here it was, sorry, my bad.
Vin, sorry for the slow reply, we normally reply immediately to query
like this but Jason is taking part in an agent competition that's
currently taking place.

If I understood the problem well, I think you can solve this with a
customised architecture for your agents (the book has a chapter on how
to do this, and there are some examples in Jason distribution). One of
the methods you can override there is the sendMsg, what you could do
there is to filter the messages to "sensor" or whatever and then send
the message using Java in the way you need.

Sorry if this has nothing to do with what you need. If I got it wrong,
let me know and I'll dedicate more time for this but not very soon,
probably only next week.

Best,

Rafael


Vin Baines wrote:

> Hi,
>
> Sorry for slow reply, I've been snowed under at work.
>
> You're right, I'm building on the work you mentioned
> (http://www.iids.org/aigaion/indexempty.php?page=actionattachment&action=open&pub_id=261&location=thesis-Rogier.pdf-bfa451fd5967ae1e6b2d3d8f990dc368.pdf).
>
>
> I'd thought it might be easier to use the existing ask performative, and
> use this in a send message, so that agents connected to a physical
> system in agentscape can report back data. e.g. send(sensor, ask, speed)
> so that any sensors currently existing in agentscape will report back
> values. Eventually I'm hoping to expand this out so that RMI type
> actions can be called on the physical sensor, using Jason to do the
> reasoning and then controlling sensors to improve, assert, disprove its
> belief base of what is going on in the real world.
>
> So, a list is maintained in agentsapce of all agents, so i can use
> Map<AgentHandle, String> to =
> lookupPublished(e.getToHandle().toString()) to lift the 'to' component
> from the previous send message, which will then give me handle(s) to all
> the sensor agents in agentscape, and pass the message on.
>
> The problem is, I can't get this message out to agentscape. I'm hitting
> the problem that Jason doesn't know that 'sensor' exists, and so does
> the sensible thing, reporting that the atom isn't valid.
>
> Is there an easy way to override this, or will it break something in the
> jason architecture? Or a better approach?
>
> Thanks for any suggestions!
>
> Vin
>
> Quoting Rafael H Bordini <r.bordini@...>:
>
>> Hi Vin,
>>
>> Jomi and I discussed this, and without knowing the details of the
>> interaction of Jason and AgentScape we can only try to guess some
>> possible alternatives:
>>
>> 1. Creating a new performative as you probably know is done completely
>> on the AgentSpeak side, but if we understand correctly you'll then need
>> to intercept this new performative and do something different with it
>> in the Java side. For this, what you need to do is to use a customised
>> AgArch class for your agents, and override the sendMsg method to use
>> the super-class method for normal messages and you handle the
>> askAgentScape in the way you want within your sendMsg method (with Java
>> code that gets the information from AgentScape and force the replies
>> into the agent mailbox, for example). There are some detail about
>> architecture customisation in the Jason book, and jason.sf.net/api will
>> probably help too.
>>
>> 2. If you have created a new Infrastructure which uses AgentScape to
>> create (and possibly distribute) agents (actually, I think someone in
>> the Netherlands did that, but I'm not sure), then you already have
>> methods for this and you'd need to change them to handle the new
>> performative.
>>
>> 3. Another possibility might be to use internal actions (which are
>> implemented in Java) to retrieve the info from the AgentScape side and,
>> e.g., add them in the agent's belief base or mailbox or however you
>> want the agent to have access to the results.
>>
>> Sorry if this isn't helpful, but we don't know much about AgentScape
>> and the details of how you interfaced it with Jason. Feel free to ask
>> further questions and send more details if you think we can help.
>>
>> Cheers,
>>
>> Rafael and Jomi
>>
>>
>>
>> Vin Baines wrote:
>>> Hi All,
>>>
>>> I'm using Jason agents in combination with the AgentScape platform,  
>>> where I've got a number of agents running in agentscape acting as  
>>> 'sensors'. Jason agents are being succesfully told information by
>>>  the  agentscape agents and executing plans based upon this info,  
>>> but I want to have the ability for Jason agents to query back to  
>>> agentscape to  verify/assert. I've been thinking about the easiest  
>>> way to do this, as a bit of java code is going to be needed to deal
>>>  with info flow  (agentscape handles & the passing of info back to  
>>> jason).
>>>
>>> The easiest (and hopefully sensible) way I can think of doing this  
>>> is  to add a new KQML performative to extend ask - e.g.  
>>> askAgentScape (for the time being - this assumes the jason agent  
>>> knows to ask agentscape rather than just ask! - but it would prove  
>>> the concept).
>>>
>>> The implementation of jason inside agentscape currently doesn't  
>>> handle any environments, so I can't just add new java code in there
>>>  to add  this extra functionality (e.g. as per room demo).
>>>
>>> So, two questions really. Where should this be added in? I'm  
>>> thinking  extend jason with the functionality using agentscape  
>>> classes, and then use the new jason.jar into agentscape so that the
>>>  asl files can use  something like askAgentScape(agName, value).
>>>
>>> Second question - does this sound a sensible way of going about it?  
>>> I'm keen to not go down a wrong hole for the 100th time..
>>>
>>> Thanks
>>>
>>> Vincent
>>>
>>>
>>> ------------------------------------------------------------------------------
>>>
>>> Crystal Reports - New Free Runtime and 30 Day Trial
>>> Check out the new simplified licensing option that enables unlimited
>>> royalty-free distribution of the report engine for externally  facing
>>> server and web deployment.
>>> http://p.sf.net/sfu/businessobjects
>>> _______________________________________________
>>> Jason-users mailing list
>>> Jason-users@...
>>> https://lists.sourceforge.net/lists/listinfo/jason-users
>
>
>

------------------------------------------------------------------------------
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day
trial. Simplify your report design, integration and deployment - and focus on
what you do best, core application coding. Discover what's new with
Crystal Reports now.  http://p.sf.net/sfu/bobj-july
_______________________________________________
Jason-users mailing list
Jason-users@...
https://lists.sourceforge.net/lists/listinfo/jason-users