Exploiting Inherent Robustness, Natural Dynamics ... - Lorentz Center

jadesoreΤεχνίτη Νοημοσύνη και Ρομποτική

13 Νοε 2013 (πριν από 4 χρόνια)

88 εμφανίσεις

Coactive Design

Why Interdependence Must Shape Autonomy

Matthew
Johnson

mjohnson@ihmc.us



Florida
Institute for

Human and Machine Cognition

Pensacola,
FL



TU Delft



December 15,
2010

Florida Institute for Human and Machine Cognition

Motivation

Problem Statement


When designing sophisticated human
-
agent systems,
what is the critical design feature of an agent or robot?



Problem Statement

Autonomy


When designing sophisticated human
-
agent systems,
what is the critical design feature of an agent or robot?



What is Autonomy?

Autonomy has two basic senses:


self
-
sufficiency,
the capability of an entity to
take care of itself.


self
-
directedness
, or freedom from outside
control.

Current Approaches



Function Allocation (
Fitts
)


characterize the general strengths and weaknesses of humans and machines


Supervisory Control (Sheridan)


a human oversees one or more autonomous systems, statically allocating tasks to them.


Adjustable Autonomy (
Dorais
)


the ability of autonomous systems to operate with dynamically varying levels of independence


Sliding Autonomy (Dias)


Same as adjustable autonomy


Adaptive Automation (Sheridan)


the system must decide at runtime which functions to automate and to what extent


Flexible autonomy (Technology horizons)


the degree of autonomous control that the system is allowed to take on, and in which this degree of
autonomy can be varied from essentially none to near or complete autonomy


Mixed
-
initiative interaction (Allen)


a flexible interaction strategy, where each agent can contribute to the task what it does best


Collaborative Control (Fong)


allows robots to benefit from human assistance during perception and cognition, and not just planning and
command generation

Self Sufficiency

Self Directedness

However…


Automated assistance of whatever kind does not
simply enhance our ability to perform the task: it
changes the nature of the task itself
1
.


Many functions in complex systems are
interdependent


1. D. A. Norman, "How might people interact with agents?," in
Software Agents
, J. M. Bradshaw, Ed. Cambridge, MA: The
AAAI Press/The MIT Press, 1997, pp. 49
-
55 (see also How might people interact with robots?
http://www.jnd.org/dn.mss/how_might_humans_int.html)
.

Autonomy

Autonomy

Problem Statement

Interdependence


When designing sophisticated human
-
agent systems,
what is the critical design feature of an agent or robot?



Coactive Design

In sophisticated human
-
agent systems,

the underlying
interdependence

of joint activity

is the critical design feature.

Disclaimer:


The authors of this work are in no way
suggesting autonomy is unimportant or not
of value. If you work in the area of
autonomy you should in no way feel
threatened or troubled by the following
discussion. If at any time you feel
uncomfortable, please inform the speaker
and he will pause and reassure you that
autonomy is important and a foundational
component to all robotic systems.

Coactive Design

In sophisticated human
-
agent systems,

the underlying
interdependence

of joint activity

is the critical design feature.

Dependence
vs.

Interdependence

In sophisticated human
-
agent systems,

the underlying
interdependence

of joint activity

is the critical design feature.

Coactive Design

Dependent Independent Interdependent

1997

2002

2008

?

In sophisticated human
-
agent systems,

the underlying
interdependence

of joint activity

is the critical design feature.

1995

Coactive Design

Dependent Independent Interdependent

What does coactive mean?

Awareness
of


Consideration
for


Capability to
support

What does coactive mean?

Awareness
of


Consideration
for


Capability to
support

…Joint Activity

What does coactive mean?

Awareness
of


Consideration
for


Capability to
support

…Joint Activity

What does coactive mean?

Awareness
of


Consideration
for


Capability to
support

…Joint Activity

Autonomy Perspective

Self
-
sufficiency

Self
-
directedness

Coactive Perspective

Self
-
sufficiency

Self
-
directedness

Current Challenges

Self
-
sufficiency

Self
-
directedness

Burden

Current Challenges

Over
-
Trusted

Self
-
sufficiency

Self
-
directedness

Burden

Current Challenges

Over
-
Trusted

Self
-
sufficiency

Self
-
directedness

Burden

Under
-
Utilized

Current Challenges

Over
-
Trusted

Self
-
sufficiency

Self
-
directedness

Burden

Under
-
Utilized

Opaque

Low to Moderate autonomy




situation
awareness

Moderate to High autonomy


-

understanding decisions

Coactive Perspective

Self
-
sufficiency

Self
-
directedness

Burden

Opaque

Opaque

Hypothesis


In human
-
agent systems engaged in joint activity, the benefits of
higher levels of autonomy cannot be realized without addressing
interdependence through coordination.

Over

Trusted

Self
-
sufficiency

Self
-
directedness

Burden

Under

Utilized

Opaque

Block World for Teams (BW4T)

Defining “Higher Levels” of Autonomy

Over

Trusted

Self
-
sufficiency

Self
-
directedness

Burden

Under

Utilized

Opaque


Did burden decrease?


Did opacity increase?


Was there an inflection point?


Did performance decrease with increasing autonomy?


What does this mean?

640
660
680
700
720
740
760
780
800
820
1
2
3
4
Time to complete task

(
in seconds)

Autonomy Level (1=low, 4=high)

Average time to complete task


vs. Autonomy Level

Prediction

Results

0%
10%
20%
30%
40%
50%
60%
70%
team 1 -low
team 1 -
high
team 2 - low
team 2 -
high
4
-
Person
Teams

Open Communication

soft
hard
Block World for Teams (BW4T)

1a

1%

1b

10%

2a

12%

2b

14%

2d

15%

3

9%

4

24%

5

0%

6a

2%

6b

4%

6c

9%

Communication in support
of

High
Interdependence

1a

2%

1b

25%

2a

3%

2b

7%

2d

7%

3

0%

4

48%

5

5%

6a

1%

6b

1%

6c

1%

Communication in
support of

Low
Interdependence

Future Directions


Does addressing interdependence improve
performance?


How does interdependence effect the results?


How do different ways to address
interdependence effect performance?


How does team size effect the result?


Does this apply to other domains?


How to we measure to costs and benefits?

We
no longer look at the problem as simply trying to make agents more
autonomous, but, in addition, we strive to make them more capable of
being interdependent.




Coactive Design

Thank You! … Questions?

Common Issues

Human Needs

Robot Needs

What

is the robot doing?

Mutual Transparency

What is

the intent of the
human?

Why did the robot do
that?

Mutual
Explainability

What did the human want
me to learn?

What is the robot going
to do next?

Mutual Predictability

What does the human
need from me?


Can we make the robot
do what we need?

Mutual
Directability

Can the human provide
help?

Does use of autonomy
add value?

Mutual Cost Benefit
Management

Will my actions provide
value to the human?