Intentional Vision in Humans and Robots

builderanthologyΤεχνίτη Νοημοσύνη και Ρομποτική

19 Οκτ 2013 (πριν από 3 χρόνια και 9 μήνες)

57 εμφανίσεις

Intentional Vision in Humans and
Robots

Dan Levin, Kazuhiko Kawamura, Megan Saylor, and Mitchell Wilkes

Vanderbilt University

On
-
line Interpretation of Intentional
and Nonintentional Behavior

Intentional Vision and Reaching
Kinematics

Machine Vision for Intentional
Behavior

Core Concepts about thinking and
representation in different systems

Intentional?


Intentional Representations:


Are closely connected to referents


Reflect beliefs and desires


Have been studied extensively in children,
but very little in adults


May be the only kind of representation
people can reason effectively about


May underlie much of higher
-
level visual
cognition

Core Concepts about thinking and
representation in different
systems: Induction in Adults


Subjects learn novel properties of a computer or
a person
-
> Does the property apply to other
things?


If subjects generalize, they consider the source
and target of the generalization to be functionally
the same.


Do subjects differentiate intentional and
nonintentional mental properties when making
inductions?

Induction Example: Intentional and
Nonintentional Mental Properties

INTENTIONAL


Imagine that John’s brain attempts to predict what
a person is going to do next using an "M
-
rule"
that takes their specific preferences into
account using an algorithm built around a set
of Gaussian kernels.


Which of the following might also use the M
-
Rule?


____ The Apple Macintosh computer

____ The Dell computer

____ OSCAR the Robot

____ The thermostat

____ The mouse

____ Popol, the adult from the Amazon

NONINTENTIONAL


Imagine that John’s brain sometimes transmits
internal information more quickly at the expense
of accuracy using a "delta rule".


Which of the following would trade off accuracy for
speed using a delta rule?


____ The Apple Macintosh computer

____ The Dell computer

____ OSCAR the Robot

____ The thermostat

____ The mouse

____ Popol, the adult from the Amazon

Results for Intentional and
Nonintentional Mental Properties

Target of Induction

% Induction

On
-
Line Behavior and
Intentionality

Jonathan Herberg, PI


Test differences in behavior during
demonstration for intentional and
nonintentional audience.


Subjects demonstrate behaviors for a
computer or a person to maximize
understanding. A picture of the
hypothetical audience audience is
visible during demo.


Results


Automatic social behavior: Subjects look to
image of audience more for Human demo.


Movements: Ratings of movement
punctuation show stronger punctuation for
computer (“Computerese”?).


Follow
-
ups to determine whether machine
vision coding of motion sequences can detect
audience, and to determine whether
anthropomorphic behavior in computer
audience affects demosntrations.

Machine Vision for Intentional
Behavior I: Video Coding

Jonathan Hunter, PI


Fuzzy logic color and texture
-
based
coding of hand and head movements.

Machine Vision for Intentional Behavior
II: 3D Position Segmentation



Kinematic Centroid segmentation and
ST
-
Isomap grouping
.

Intentional Vision and
Reaching


Pilot data demonstrating impact of intentional
model on reaching kinematics.


Creation of teleoperated robotic head to
simulate mechanical intentional system.


Goals for Coming Year


Follow
-
up on core concepts findings


Complete initial reaching experiments


Integrate segmentation data and
machine vision analyses


Lay foundations for bootstrapping
machine vision segmentation into
intentional analysis