Cognitive Robotics

 

Mental Simulation in Robots

Page history last edited by Nicholas Davis 1 yr ago

Integrating cognition, perception and action through mental simulation in robots

 

Back to Research Findings

 

Research Notes:

(text blocks quoted from article)

 

 

"These include the problems involved in

 

data fusion, symbol grounding and flexibly combining

 

reasoning, planning, perception and action."

 

1)       Data fusion:

a.       Combining data from multiple modalities of the same object. Different sensors are getting different data of the same object. The robot has to decide that it is indeed the same object.

2)       Knowledge Representation:

a.       Different knowledge is best represeneted by different theoretical systems. In order to have a robust understanding, the robot would have to employ multiple systems.

b.      Behaviour based and reactive research have handled this situation by not making representation that deal with past and future states. The robot only takes the input and acts on it according to certain algorithms. However, this limits the way the robot can interact with the environment.

                                                               i.      This is basically functional semantics. The robots does not know it is in a space, it is only receiving data and reacting, no planning, no goals or remembrance of past events.

 

"Polyscheme differs from traditional

 

cognitive architectures based on one or a few data

 

structures by enabling inference based on multiple

 

data structures. Polyscheme differs from the many

 

multi-agent system architectures that encapsulate specialized

 

algorithms in modules by enabling every step

 

of every algorithm to be executed using multiple representations

 

and be potentially assisted by every other"

 

3)       Poly bot

a.       Using a system of integrating many different inferential and knowledge representation systems by being able to run mental simulations that are based upon perceptual components.

b.      It uses the paradigm of functional semantics of only reacting to the current state of the world, but the world is a constructed, mental simulation of the world, so inferences may be drawn during this simulation and implementation of ‘reactives’

 

"Polyscheme takes “reactive” components that

 

typically choose their actions only with reference to the

 

currently sensed state of the world, and allows them to

 

“react to” represented or simulated states of the world.

 

Using strategies for choosing which simulations to run

 

(i.e., what time, place or hypothetical world to simulate),

 

Polyscheme implements high-levelAI algorithms

 

by executing mental simulations. Because simulations

 

are composed in part of reactive and perceptual subcomponents,

 

reasoning is constantly and thoroughly

 

integrated with perception and action."

 

4)       Perception and mobility modules called specialists are all modules that deal with executing action or perceiving the environment in one specific manner. for example there is a color segmentation specialist part of the polyscheme.

a.       Specialists communicate with each other, for example the perception specialist communicates with the mobility specialisit. Perception sees a cone at time now and tells this to the mobility, who knows to avoid objects.

b.      Specialists deal with things in a basic propositional language that is common across all specialists. Sometimes the truth value of a certain proposition that a specialists holds will change and the other specialists need to be informed of this.

5)       Each specialist uses the information it has to run mental simulation of the data pertaining to its domain of inquiry. The language that it uses in this domain does not have to be common to all specialists, but the output needs to be translated into the common language, so the other specialists understand.

 

"Specialists are not committed to using these

 

units and these predicates in their own internal computations

 

or even using representations based on predicates

 

on objects. They must simply be able to translate

 

between this representation and their own internal representation.

 

By separating the representation used for

 

communication from the representation used for inference,

 

we achieve the benefits of a common ontology

 

without the rigidity often associated with such knowledge

 

representation schemes."

 

6)       Any state that is not currently present is referred to as non-immediate.

 

"To represent non-immediate situations, Polyscheme’s

 

ontology includes temporal intervals and hypothetical

 

worlds. The last two arguments of every proposition

 

input to and output from specialists are, respectively,

 

a time and world argument. For example, the

 

proposition that object, o, is located at point p, in

 

hypothetical world, w, at time t is indicated: Location(

 

o,p,t,w)."

 

7)       Memory serves to decide truth value about past states and causal structures serve to indicate truth value about future states. For example, the causal rule specialists will predict that a ball rolling behind an object will continue to exist behind that object even though it is occluded from view.

a.       The term simulus is used to describe inputs to specialists from a simulated world.

8)       All specialists have to work on the same proposition at the same time because many different specialists are required to create a robust understanding of the situation.

 

"• Polyscheme calls StanceOn(P) for each specialist

 

to determine the consensus truth value of all the

 

specialists on P.

 

For each specialist, Polyscheme calls the function

 

ReportOpinion(P,specialist,tv) to report

 

specialists’ truth values for P to each other.

 

Polyscheme calls RequestedFoci( ) to get

 

propositions the specialists would like to focus on soon."

 

 

9)       A simulated world results from the regulator, polyscheme, the kind of governor who is in charge of all of the specialist, who calls for their attention and asks what their opinion/stance is on a given proposition, changes the time value and world value in the proposition. If it gives a proposition with a time value of future, then all of the specialists will work together to provide each other with information that they can draw inferences on and pass information along to each other, communicate in order to create  a coherent simulation of the world.

 

 

"When Polyscheme

 

focuses on a particular time and world, the architecture

 

forces all specialists to focus on that time and world.

 

The sum effect of this attention will be that specialists

 

will make inferences about that world and therefore

 

elaborate Polyscheme’s representation of it. Since

 

some of these inferences will involve the consequences

 

of states and events in this world, Polyscheme will focus

 

on subsequent times in the same world."

 

 

10)   The attention of the specialist is guided by polyscheme by their requestedfoci() function. Polyscheme uses the focus  manager chooses from the possible range of activites based upon the level of urgency specialist.

 

 

"Polyscheme thus continuously chooses a proposition

 

to focus on, allows specialists to communicate

 

about and make inferences about this proposition and

 

then chooses the next proposition to focus on based on

 

specialists’ requests. Through their ability to request

 

Polybot to focus on a proposition, specialists can influence

 

the flow of attention and hence computation."

 

 

Comments (0)

You don't have permission to comment on this page.