opencog-dev team mailing list archive

[Ben Goertzel <ben@xxxxxxxxxxxx>]

By popular demand, I'm forwarding this email to the list...

ben g

---------- Forwarded message ----------
From: Samir Araújo <samir@xxxxxxxxxxxxx>
Date: 2009/8/10
Subject: Coding an OpenCog implication for the robot to carry out a command
like "Do Tai Chi"
To: Troy Huang <huangdeheng@xxxxxxxxx>
Cc: Ben Goertzel <ben@xxxxxxxxxxxx>


 Hi Troy,

in order to convert a sentence into a command, you will need to create an
ImplicationLink that has in its preconditions a test to check the presence
of the desired sentence into the AtomTable.
i.e.

ImplicationLink
     EvaluationLink *(preconditions. You can use an AndLink to evaluate more
than one precondition)*
         PredicateNode:"actionDone"
             ListLink
                 ExecutionLink
                     GroundedSchemaNode:"say"
                     ListLink
                           AvatarNode:"RobotOwner"
                           SentenceNode:"to:robot_id: *do tai chi*"

     ExecutionLink *(effect)*
         GroundedSchemaNode "tai_chi"
         ListLink

(You can write ImpLinks directly in C++ or using a bind script language like
Scheme)

You must to pass, as an argument, the Handle of the defined ImpLink to the
pattern match (opencog/query/PatternMatch::imply(Handle)).If the
preconditions are satisfied, the PatternMatch will create the effect  (in
this case an ExecutionLink containing the schema that will execute the
desired action) into the AtomTable and then it will return the handle of the
execution link. Finally, you must to evaluate the returned ExecutionLink and
then call the appropriate library to make the robot execute the action.  If
you're implementing the action procedure using Combo, for example, you will
need to define a built-in Combo procedure (written in C++) and send the
GroundedSchemaNode to Combo Procedure Interpreter to be executed. This
process must be repeated every time a new sentence was said to the robot.

Well, this is a very simplified description of the whole process. Obviously
you will need to build an ImpLink that detects not only the presence of a
SentenceNode into the AtomTable, but it must to detect what was the latest
said sentence, who said, etc... So, if you send me more informations  about
how you're integrating OpenCog with the robot, I can help you to prepare the
module that will evaluate the rules (ImpLinks) and execute the actions in
the robot. (are you using Embodiment? are you using Relex to parse the said
sentences? if you're not using relex, how the sentence comes from the proxy?
etc...)

Samir












-- 
Ben Goertzel, PhD
CEO, Novamente LLC and Biomind LLC
Director of Research, SIAI
ben@xxxxxxxxxxxx

"Truth is a pathless land" -- Jiddu Krishnamurti