[ACT-R-users] ACT-R for simulation purposes

[db30 at andrew.cmu.edu]

--On Tuesday, February 22, 2011 11:40 AM +0100 Ana Maria Navarro Cerda 
<annacer at itaca.upv.es> wrote:


> However, there are some issues that remain unclear to me:
>
> ·         These simulation models  we are developing (task models) are
> not fixed, we have a repository of tasks and the developers can
> "compose" the final simulation model as a sequence of tasks, allowing to
> have an infinite number of simulation of models. So we would need that
> Act-R allowed certain modularity and reuse of the code for the models.
> Is this feasible? Is there any modification of ACT-R allowing this?
>


>From a technical standpoint, there is nothing that prevents composing a 
model
from several smaller components.  An ACT-R model is essentially just Lisp 
source
code and thus one could load individual pieces as necessary.  From a 
practical
standpoint however there would of course be software engineering type 
issues
with respect to creating those components such that they can be used 
together.
There's nothing in ACT-R which specifically addresses issues like that and 
the
modeler wishing to do so would be responsible for creating and maintaining
compatible and non-conflicting representations and production sets as 
desired.


> ·         Another question is concerning the simulation: if there are
> external events happening in the simulation (ie: a car in front of us
> when driving),  should these "events"  also be part of the declarative
> memory?

The answer to that question depends on how you're looking to model the
task.  ACT-R has perceptual modules for vision and hearing which can
be used to situate a model in a world.  Thus, one could use those to
provide the input to the model in a more natural way.  To do that one
has to create what we call a "device" for the model which converts the
simulation output into an appropriate form for the model to use.  The
device is also responsible for converting the model's motor and vocal
output into the proper actions needed by the simulation environment.
Alternatively, it is possible to forgo the perceptual and motor components
of ACT-R and simulate a world for the model through the manipulation of
chunks in declarative memory as well as the goal and imaginal modules.
Which method you choose would depend on your objectives for the modeling
effort.

> If the answer is yes, do we have APIs that allow us to integrate
> this "events" into chunks in an automatic way?
>

There is a reference manual which provides the API for the ACT-R system
itself, and there is documentation and example code included with the 
ACT-R
sources for creating devices for a model.  If one chooses to not use a 
device
there is no "one way" to go about simulating an environment for the model,
and thus one would have to decide on how to do that essentially from 
scratch.

> ·         Does it exist t any open repository of ACT-R tasks models that
> we could use?
>
>

There is no repository of ACT-R models, but models for some published work
are available through the ACT-R web site: 
<http://act-r.psy.cmu.edu/models/>
and other researchers have made models available in other ways as well.
However, some of those are models which were written for older versions of
ACT-R which won't work properly as-is in the current software, and no two 
would
likely be useable together without modification because they wouldn't have
been designed for that purpose.  So, there is not a library of ready to 
use
component models which you could just download and use.


>
> I would really appreciate any advise you could give me. This is a new
> world for me, and it's difficult to go into it without getting lost.
>

The first thing I would suggest is to download the ACT-R software and work
through the included tutorial materials from 
<http://act-r.psy.cmu.edu/actr6/>.
That will introduce all of the basic components of the system and give you 
a
feel for what creating an ACT-R model is like.  Along with that, if you are
not already comfortable with Lisp programming you might also want to get 
some
instructional materials for that as well because creating an interface for
the model (whether a device or direct chunk manipulation) is going to 
involve
Lisp programming.  After that, you should be in a better position to decide
whether ACT-R is the tool you want to use and how you may want to go about
using it.

Hope that helps, and if you have any questions or problems about the ACT-R 
software
specifically, feel free to email me about them.

Dan