Download - CERA-CRANIUM: A Test Bed for Machine Consciousness Research

Raúl Arrabales, Agapito Ledezma and Araceli Sanchis

Computer Science DepartmentCarlos III University of Madrid

http://Conscious-Robots.com/Raul

International Workshop on Machine Consciousness 2009PolyU, Hong Kong, 14th June 2009

CERA-CRANIUM: A Test Bed for Machine Consciousness

Research

http://conscious-robots.com/Raul

http://scalab.uc3m.es/

http://www.conscious-robots.com/index.php

www.Conscious-Robots.com

Contents

Introduction and objectives. Related work. Objectives. CERA-CRANIUM. Experimentation settings. Example of application. Conclusions.

2


Introduction (I) Theories of consciousness

Philosophical or psychological background.

Metaphorical descriptions.

Need to bridge the gap between theories and implementation.

3


Introduction (II) Cognitive theories of consciousness

Global Workspace Theory (Baars, 1997).

Multiple Draft Model (Dennett, 1991).

Inspiration for the design of a partial computational model of consciousness.

4


Introduction (III) Common ground

Non-unitary mechanisms producing the unity of self.

In other words, conscious contents emerge as a result of competition/collaboration (Minsky, Dennett, Hofstadter, Baars, Shanon).

5


Introduction (IV) Objectives

To understand how cognitive skills associated with consciousness can be integrated effectively.

To test different machine consciousness models.

6


Introduction (V) Functionalism:

Is this a reductionist approach? Does this mean that phenomenal

aspects are ignored?

7


Related work (I) Shanahan’s (2005, 2006) cognitive

architecture.

IDA and LIDA (Ramamurthy et al., 2006).

Computational Agent Framework for Consciousness (Moura & Bonzon, 2004).

CERA-CRANIUM (Arrabales et al., 2007, 2008).

8


Related work (II) Common denominator:

Shared workspace and specialized processors.

Different approaches in terms of: Architecture. Problem domain. Perceptual flow. Decision taking. Modulation.

9


Objectives Experimentation platform.

Test bed for high level cognitive approaches.

Generic but configurable cognitive architecture.

10


CERA-CRANIUM (I) CERA-CRANIUM Provides:

Mechanisms for specialized processors

Creation, Association, Combination, Competition.

Mechanisms to regulate the former processes.

11


CERA-CRANIUM (II) CERA: layered control architecture.

CRANIUM: runtime tool for the creation and management of high amounts of parallel processes in shared workspaces.

12


CERA-CRANIUM (III) CERA: layered design

Sensorimotor services layer. Physical layer. Mission-specific layer. Core layer.

13


CERA-CRANIUM (IV)

14

Agent

Sensors

Actuators

CERA World

Accessible environment

Reachable environment

Sensor Services

Motor Services

Physical Layer

Mission-specific Layer

Core Layer

Different levels of description


CERA-CRANIUM (V) CRANIUM

Blackboard Pandemoniu

m

15

SpecializedProcessors

Working Memory

Focus of Attention

Contexts

Interim Coalition


CERA-CRANIUM (VI) Perceptual flow:

Sensory data. Single percepts. Complex percepts. Mission percepts.

16


CERA-CRANIUM (VIbis)

17

Sensor

Single Percepts

Sensor Preprocessors

Sensor Readings

Timer

Proprioception

Sensor Service

Agent

CERA sensory-motor services

CERA physical layer

Percept Aggregators

S (World)

eventsδSj N(δSj)N(δSJ)

j t

Complex Percept

M(SCJ)

Knowledge representation


CERA-CRANIUM (VII) Behavior generation:

Mission behaviors. Simple behaviors. Single actions. Motor controller commands.

18


CERA-CRANIUM (VIIbis)

19

Actuator

Atomic Actions

Action Preprocessors

Single Actions

Timer

Proprioception

Motor Service

Agent

CERA sensory-motor services

CERA physical layer

Action Planners

S (World)

ActionδBi

N(δBi) N(δBI)j t

Simple Behaviors

Action Dispatcher M(BCI)

Action generation


CERA-CRANIUM (VIII) Bottom-up flow

20

CERA Core Layer

CERA M-S LayerCERA Physical LayerSensor Service

Sensor Service

Single Percepts

…

CRANIUM Workspace

Complex Percepts

…

CRANIUM Workspace

Mission Percepts

Sensor Preprocessors

…Specialized Processors

Sensor Service

CERA S-MSensors

…Percept

Aggregators


CERA-CRANIUM (IX) Top-down flow

21

CERA Core Layer

CERA M-S LayerCERA Physical LayerMotor Service

Motor Service

Single Actions

…

CRANIUM Workspace

Simple Behaviors

…

CRANIUM Workspace

Mission Behaviors

Action Preprocessors

…Specialized Processors

Motor Service

CERA S-MActuators

…Action

Planners


CERA-CRANIUM (X) CRANIUM processor types

Sensor preprocessors Raw sensory data single

percepts

22


CERA-CRANIUM (XI) CRANIUM processor types

Action preprocessors Atomic actions Single actions.

23


CERA-CRANIUM (XII) CRANIUM processor types

Percept aggregators Single percepts complex

percepts. Complex percepts complex

percepts.

24


CERA-CRANIUM (XIII) CRANIUM processor types

Reactive processors Single percept simple

behavior. Complex percept simple

behavior.

25


CERA-CRANIUM (XIV) CRANIUM processor types

Action planners Simple behavior atomic

actions.

26


CERA-CRANIUM (XV) CRANIUM processor types

Sensory predictors Single percepts mismatch

complex percept. Complex percepts mismatch

complex percept.

27


CERA-CRANIUM (XVI) Multi-level concurrent feedback loops

28

CERA M-S Layer CERA Core LayerCERA Physical LayerCERA S-MWORLD

(a)

(b)

(c)


CERA-CRANIUM (XVII) Physical level feedback loops

29

Contact Sensor Service

Drive Service

Single Percepts

Simple Behavior

MotorControllers

Crash!

Contact Sensors

WORLD AGENT CERA S-M

CERA Physical Layer

Actions

Complex Percept Reactive

Processor

Workspace

Action Planner

Percept Aggregator


CERA-CRANIUM (XVIII) Software architecture

30

Agent sensory-motor machinery

.Net FrameworkCCR

CRANIUM

DSS

CERA Physical

CERA Sensory- Motor Services

.Net FrameworkCCR

CRANIUMDSS

CERA Mission-specific

.Net FrameworkCCR

DSS

CERA Core

MCCM Configuration

DSSP DSSP


CERA-CRANIUM (XVIIIbis) Software architecture

31

Single Percepts

Complex Percepts

…

Workspace

…

Workspace

Mission Percepts

Core Layer

Physical Layer Mission-Specific Layer

Modulation Commands


CERA-CRANIUM (XIX) The proposed cognitive

architecture is focused on: Selecting next content of conscious

perception. Selecting next action to be

executed.

32


Experimentation settings Variables:

Cognitive model of consciousness. Agent (physical or simulated). Problem domain and mission. Environment (physical or

simulated).

33


Applications

Current applications:

34


Application example (I)

Autonomous exploration and mapping

35


Application example (II)

Autonomous exploration and mapping.

36


Application example (III)

Percepts (bumpers).

37

(0,0)

-52º

-19º0º

19º

52º

BAb5

BR

b1

b2

b3 b4

b5


Application example (IV)

Single percept (bumper).

38

Impact

j referentLeft-j referent

Right-j referent

N(δSJ)


Application example (V)

Complex percept (bumper).

39

Impact

Left-j referent

Right-j referent

j referent

M(SCJ)


Application example (VI)

CERA Core Layer Design.

40

Physical Layer

Mission Percepts,Complex percepts,Mismatch Percepts,Novelty Percepts,

…

M(SCJ)

…

CRANIUM Workspace

…

CRANIUM Workspace

M-S Layer Core Layer

Workspace Commands

Current Model State

Contextual J-Index

Calculation

Cognitive Model

Meta-goalsRule-based

system


Conclusions (I)

From sensory to qualia?

41

Sonar transducer

Ultrasonic beam (three-dimensional cone)

Left-j referent

Right-j referent

j referent vector(| j | = range measurement)


Conclusions (II)

Different implementations of CERA layers.

Explore workspace modulation techniques.

42

www.Conscious-Robots.com 43

Thank you

Download - CERA-CRANIUM: A Test Bed for Machine Consciousness Research

Top Related