Transcript Introduction of Artificial Moral Agent Incorporating Soar and

Introduction of Artificial
Moral Agent in
Relation with Soar
Jong-Wook Kim, Chien Van Dang
Dept. Electronic Engineering
Dong-A Univ., Busan, Korea
June 10, 2016
36th Soar Workshop
1
Contents

Robot Ethics

Artificial Moral Agent

My Research Experience

•
Humanoid Robotics
•
Global Optimization
Conclusion
2
Why Ethics?


Being moral is the only feature that human has among the animals.
Human is coexisting with increasing robots








Social robots
Telepresence robots
Collaboration robots in factories
Humanoid robots in disaster sites
Military robots
Self-driving cars
Drones
Human easily personifies robots.
High expectation as semi-human
 Empathy
 Excessive fear


AI without morality can be a disaster
3
Human














Speculation with Philosophy and Ontology
Introspection and Reflection
Curious and Creative
Pursuit of Truth, Liberty, Justice, Equality…
Conscience and Morality
Various emotions of Joy, Anger, Love, Hatred…
Basic instinct to eat, love, fight…
Acquired desire for Power, Honor, Possession…
Social being living in community
Hand and Tools
Languages and letters
Enjoy Art, Sports, Habits…
Ego and Free will
Prejudice
4
Relevant Questions






Should a robot never lie? This can cause a serious problem if the enemy
captured it.
Is there any danger with emotional attachment to robots?
If a robot or driverless car has decided to kill one person to save 100
people in an emergency situation, is it acceptable to humans? And what
if the one to be killed is the owner of the robot?
If a robot happen to give a harm to a child passing by, who is
responsible for this accident?
Is a robot deserved to be respected as another being by human?
More questions case by case…
5
Robot Ethics

Definitions of Robot Ethics
Professional ethics of roboticists
 Moral code programmed into the robots themselves
 Self-conscious ability to do ethical reasoning by robots


Approach of Morality

Top-down, rule-based approach (Asimov’s Three Laws of Robotics)
 Deontologist: One must intend to obey the rules even if the
consequences will be bad (I. Kant).
 Utilitarian: The main and only rule is always to make the future
consequences as good as possible (J. S. Mill).

Bottom-up, art/science of living a good life


Trial-and-error learning of what constitutes (un)acceptable behavior for
a good or bad robot
Virtue ethics (Aristotle)


“What should I be?” instead of “What should I do?”
Virtues are dispositions to act in a certain way (would-be habits).
6
Research Trend on RoboEthics
MedEthEx (Michael & Susan Anderson): Medical and Ethical Expert
System
Robot Ethics sessions and papers in IEEE RAS Conference (ICRA2011,
IACAP2011, AISB 2012) and ICSR
Generate Acceptance/Rejection using DIARC/ADE cognitive
architecture, Gordon Briggs, Tuffs Univ.
7
Overview of AMA Project
Project Goal
- Development of Artificial Moral Agent integrating Soar and ROS
referring to morality of 10-years-old human level
- Application of AMA to social/care robots
- Establishment of Moral Turing Test with suitable scenarios
- Preparation of Functional Morality for certification of commercial
robots
Roadmap
Short
Conversation
Robot
Protection
Image
Processing
Emotional
Response
Voice
Recognition
Top-Down
Ethical Layer
ROS
Emotion
Recognition &
Generation
Soar
Bottom-Up
Ethical Layer
Robot
Etiquette
Manipulation
Ethical
Behavior
SLAM
Walking
/Moving
Objection to
Immoral Directive
8
Project Plan
1. Theoretic background on Roboethics for AMA (Year 1~3)
Seoul National Univ. of Education
Development of Top-Down ethical inference and decision system
with ethical modules referring to Kohlberg’s pre-conventional level
Development of Bottom-up moral learning system with investigation
and analysis of the current status and ethical problems of social/care
robots in use
Analysis of realistic moral cases collected from health care
employees and validation of ethical modules from it
Presentation of dilemma and situation analysis for Moral Turing Test
Modification and Upgrade of ethical modules by feedback from
development processes
9
Project Plan
2. AMA Software Development (Year 1~3)
Dong-A Univ.
ROS-Soar package for AMA by connecting ROS and Soar
Integrating speech recognition and image processing to ROS-Soar
package
Top-Down ethical inference layer and Bottom-Up Moral layer to
ROS-Soar package
Connection of IoT network (ROCON) and ROS-Soar package
Integrating cloud system (Rapyuta) with ROS-Soar package
Moral context awareness package
Simulation of human-robot coexistence in
Gazebo and execution of MTT with success
10
Project Plan
3. Development of AMA Robot (Year 4~5)
Seoul National Univ. of Education & Dong-A Univ.
Task and Motion Intelligence with HRI





Pick and place task of various objects
Errand tasks with SLAM for beverage or food
Face, voice, emotion recognition of the companion people
Generation and expression of robot emotions
Short conversation with human
Moral Decision Capability for 10-Years-old Human Level
 Refer to human etiquette
 Appropriate emotional response to a person in abnormal emotion or
mood
 Responding to multiple people in a proper way based on their priorities
 Identification and objection to Immoral Directive for other people or
robot itself
11
My Research Experience: Humanoid Robot

3D Modelling of legs
DH
Matrix
i
B1
1 
di

ai
i
Aux.
matrix
0
0

2
-
0
l1
0
-
0
-
B2
2

 1 
2
B3
 2
0
l2
B4
 3
0
0


B5
2
0
l3
0
B6
1
0
l4
0
B7
2
B8
3 
B9
-
2

0
1

0

0
0
0
1
0
1
0
0
0
1
0

0

0
0
1
0
0
0
0
1
0
0
0

0
0

 4  
0
l6
0
-
B10
 5
0
l7
0
-
B11
 6
0
0
B12
4
0
l8

2
2

2
0
0
0

1
0
l5 
0

1
-

2
0
-
12
Humanoid Modelling

3D Modelling of upper body
DH
Matrix
i
di
ai
0.016
B14
 7
0.024
B15
5  
0
l9
B16
 8
0
l10
B18
 9
-0.024
0.016
B19
6  
0
l11
B20
 10
0
l12
B22
3
l13
0
0
l14
B23
 11 

2
i

2


2
0


2

2
0

2
0
13
Modelling with Projection Based Method
X 1  X 0  l1s C1 , Y1  Y0  l1c S1 , Z1  Z 0  l1s S1 ,


X 2  X 1  l2sC12
, Y2  Y1  l2c S1 , Z 2  Z1  l2s S12
,




X 3  X 2  l4s C123
, Y3  Y2  l4c sin  12  , Z 3  Z 2  l4s S123
,
2




X 4  X 3  l6s C1234
, Y4  Y3  l6c S123
, Z 4  Z 3  l6s S1234
,



X 5  X 4  l7s C12345
, Y5  Y4  l7c S123
, Z 5  Z 4  l7s S12345
,



X 6  X 5  l8s C123456
, Y6  Y5  l8c S1234
, Z 6  Z 5  l8s S123456
14
Modelling with Projection Based Method
l shc
l shc 


X 7  X n , Y7  Yn  sin 12  , Z 7  Z n  S12 ,
2
2
2




X 8  X 7  l9s C1237
, Y8  Y7  l9c S125
, Z 8  Z 7  l9s S1237
,



X 9  X 8  l10s C12378
, Y9  Y8  l10c S125
, Z 9  Z 8  l10s S12378
,
lshc
lshc 


X 10  X n , Y10  Yn  sin 12  , Z10  Z n  S12 ,
2
2
2




X 11  X 10  l11s C1239
, Y11  Y10  l11c S126
, Z11  Z10  l11s S1239
,

c 
s 
X12  X11  l12s C1239
10 , Y12  Y11  l12 S126 , Z12  Z11  l12 S123910 ,



X 13  X 12  l13s C123
, Y13  Yn  l13c S12
, Z13  Z12  l13s S123
,

c 
s 
X14  X13  l14s C123
11, Y14  Y13  l14 S12 , Z14  Z13  l14 S12311.





l1s  l1 C1 , l2s  l2 C1 , l4s  l4 cos 12   , l6s  l6 C123
, l7s  l7 C123
, l8s  l8 C1234
2




l1c  l1 S1 , l2c  l2 S12 , l4c  l4 , l6c  l6 S1234
, l7c  l7 S12345
, l8c  l8 S123456
15
Optimal Biped Walking
Biped walking pattern is generated including
Particle swarm optimization method
 Reference ZMP trajectory
 Desired step length and swing foot trajectory

Swing foot trajectory
0.016
actual
reference
0.014
0.012
z (m)
0.01
0.008
0.006
0.5
0.004
0.002
0
-0.06
0.4
0.3
-0.04
-0.02
0
x (m)
0.02
0.04
0.06
actual ZMP
reference ZMP
0.05
Z (m)
0.2
0
Y (m)

0.1
-0.05
0
-0.2
0.2
0
0
-0.08
0.2
-0.2
-0.06
-0.04
-0.02
0
X (m)
0.02
0.04
0.06
0.08
Y (m)
X (m)
16
Motion Control
Sensory Reflex Control
X
 1   l ft d ZMP

 l2s cos1   2  

  1   2 
 2  k
cos 




l2s




( hp ) X
 3  k ZMP
d ZMP
( kn )
ZMP
avg-zmp-x
prev-avg-zmp-x
1
0.8
normalized ZMP
x
1.2
left foot
right foot
1
0.8
0.6
0.4
0.2
x
normalized ZMP

0.6
0
0.4
-0.2
0.2
0
5
10
15
20
25
time (s)
0
-0.2
0
1
2
3
4
time (s)
5
6
7
8
17
Walking Control

Walking on Inclined Floor

Sagittal joint angles in the leg
 tr
ˆtr  ˆ  ˆ1  ˆ2  ˆ3
3
~
tr  ˆ  ˆ1  1  ˆ2  ˆ3  ˆtr  1


~ ˆ 
~
1  k( an) tr  ˆtr
3  k
( hp)

tr
Yc , Z c 
 2
Y 2 , Z 2 
 1
Z
tr
4
Coronal joint angles in the leg

Y
ˆtr  ˆ  ˆ1  ˆ2


~  ˆ 
~
1  k( an) tr  ˆtr
2  k( hp)
tr
tr
18
Global Optimization


Global optimization is crucial for real time motion generation and
intelligent reasoning of humanoid robot.
Dynamic Encoding Algorithm for Searches (DEAS)
0
(1/4)
BSS
00
(1/8)
000
(1/16)
UDS
01
(3/8)
001
(3/16)
1
(3/4)
010
(5/16)
011
(7/16)
100
(9/16)
10
(5/8)
11
(7/8)
101
(11/16)
110
(13/16)
111
(15/16)
0000 0001 0010 0011 0100 0101 0110 0111 1000 1001 1010 1011 1100 1101 1110 1111
(1/32) (3/32) (5/32) (7/32) (9/32) (11/32) (13/32)(15/32)(17/32)(19/32)(21/32)(23/32)(25/32)(27/32)(29/32)(31/32)
SDE
EA
MLSL
IA
TUN
TS
BR
2700
430
206
1354
-
CA
10822
-
-
326
GP
5439
460
148
RA
-
2048
SH
241215
H3
H6
DEAS
eDEAS
uDEAS
492
94
77
1469
-
87
74
-
-
486
103
113
-
-
-
540
304
273
-
-
7424
12160
727
137
268
3416
-
197
-
-
508
189
131
-
-
-
-
-
2845
1760
261
19
19
Global Optimization

Genetic Algorithm, Particle Swarm Optimization, Mesh Adaptive Direct
Search, Interstellar Search Method have been hybridized, developed,
and applied to optimal design problems including motor core design and
pattern generation of biped walking.
Genetic Algorithm
Particle Swarm
Optimization
Mesh Adaptive Direct Search
Interstellar Search Method
20
Conclusion





Significance of robot ethics is addressed.
Artificial Moral Agent Project newly launched in Korea is
introduced with brief plans.
Walking and motion control of Humanoid Robots are briefly
explained.
Global optimization method are now integrated and
innovated by novel ideas.
Soar will be used as core agent architecture in AMA and
real-time intelligence and control of humanoid robots.
21
Q and A
Thanks for Attention.
22