Introduction of Artificial Moral Agent in Relation with

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 36 th Soar Workshop 1

Contents l Robot Ethics l Artificial Moral Agent l My Research Experience l • Humanoid Robotics • Global Optimization Conclusion 2

Why Ethics? Ø Ø Being moral is the only feature that human has among the animals. Human is coexisting with increasing robots l l l l Ø 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 l Empathy l Excessive fear l Ø AI without morality can be a disaster 3

Human Ø Ø Ø Ø 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 Tools Languages and letters Enjoy Art, Sports, Habits… Ego and Free will Prejudice 4

Relevant Questions Ø Ø Ø 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 Ø Definitions of Robot Ethics Professional ethics of roboticists l Moral code programmed into the robots themselves l Self-conscious ability to do ethical reasoning by robots l Ø Approach of Morality l Top-down, rule-based approach (Asimov’s Three Laws of Robotics) n Deontologist: One must intend to obey the rules even if the consequences will be bad (I. Kant). n Utilitarian: The main and only rule is always to make the future consequences as good as possible (J. S. Mill). l Bottom-up, art/science of living a good life n l Trial-and-error learning of what constitutes (un)acceptable behavior for a good or bad robot Virtue ethics (Aristotle) n n “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 Robo. Ethics Med. Eth. Ex (Michael & Susan Anderson): Medical and Ethical Expert System Robot Ethics sessions and papers in IEEE RAS Conference (ICRA 2011, IACAP 2011, AISB 2012) and ICSR Generate Acceptance/Rejection using DIARC/ADE cognitive architecture, Gordon Briggs, Tuffs Univ. 7

Overview of AMA 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 Io. T 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 Ø 3 D Modelling of legs DH Matrix Aux. matrix 0 0 - 0 0 0 0 - - 0 0 0 12

Humanoid Modelling Ø 3 D Modelling of upper body DH Matrix 0. 024 0. 016 0 0 -0. 024 0 0. 016 0 0 0 13

Modelling with Projection Based Method 14

Modelling with Projection Based Method 15

Optimal Biped Walking Ø Biped walking pattern is generated including Particle swarm optimization method l Reference ZMP trajectory l Desired step length and swing foot trajectory l 16

Motion Control Ø Sensory Reflex Control 17

Walking Control Ø Walking on Inclined Floor l Sagittal joint angles in the leg l Coronal joint angles in the leg 18

Global Optimization Ø Ø Global optimization is crucial for real time motion generation and intelligent reasoning of humanoid robot. Dynamic Encoding Algorithm for Searches (DEAS) SDE EA MLSL IA TUN TS BR 2700 430 206 1354 - CA 10822 - - 326 GP 5439 460 148 RA - 2048 SH 241215 H 3 H 6 DEAS e. DEAS u. DEAS 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 Ø 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 Ø Ø Ø 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