09kuniyoshi-eng.ppt

1,325 views
1,182 views

Published on

0 Comments
1 Like
Statistics
Notes
  • Be the first to comment

No Downloads
Views
Total views
1,325
On SlideShare
0
From Embeds
0
Number of Embeds
3
Actions
Shares
0
Downloads
10
Comments
0
Likes
1
Embeds 0
No embeds

No notes for slide
  • シラード:連鎖反応の発明.原子炉の特許.アインシュタイン-ルーズベルト書簡の推進,執筆者.後に原爆投下反対.
  • スタティック(静的)なタスクでは様々な設計解がありうる.ダイナミック(動的)な運動では設計解の幅が狭い. 衝突への応答で身体(ハードウェア)が重要になる.実機を開発する理由にもなる. ソフト的な制御の限界.ハードを利用する方が妥当.
  • 日常の動作に有利な働き. ハムストリング・・・前屈すると腿の裏がつっぱります.ひざを曲げておけば大丈夫です.
  • 起き上がり動作を考える上で注目すべき点は次のようになります. 1 つは,瞬発力が必要な動作であり,タイミングが重要となると考えられます. このような動きはスポーツなどでよく見られる動きです. 2 つ目は,転がり時には地面と複雑な接触があり,地面の特性,体の幾何形状,重量,慣性などの 力学的性質に従うため,姿勢の制御が困難です. 3 つ目は,着地時の姿勢とエネルギー状態は非常にクリティカルな条件を満たさなければなりません. つまり,起き上がりの問題は姿勢の制御が困難な状況下において,ある目的の状態に移行するという問題といえます。 我々は,タイミングなどの「コツ」のようなものを満たすことでこのようなことを行う制御を 提案できないかと考えています.( 50 秒)
  • つぎに , われわれは , 重心を足の上に乗せつづけることを実現しました . 足の上に重心を乗せたまま安定させるためにはどうしても , 姿勢センサからのフィードバックが必要となりました . 現在は足の裏にセンサがないため , 頭についている角速度計の値をもとに足首や腰を制御することで 安定性を保っています . ( 50 秒)
  • 用いる触覚センサは、去年 RSJ で発表した、このようなセンサです。 用いるセンサは反射型フォトインタラプタの上に発泡ウレタンをおいた簡単な構成のセンサで ウレタンの変形によって、光の拡散領域が変化することを利用しています。 また、これらをモジュール化し、複数のセンサと通信機能をもつフレキシブルセンサを開発しました。 形状を工夫することで、柔軟に曲面に実装可能です。
  • 実際に、ヒューマノイドに実装しました。 背中には、 308 点、胸側には 192 点、でん部には 150 点、股間は 62 点、 足裏には 56 点実装しております。
  • 実際に実現した起き上がり動作です。
  • 09kuniyoshi-eng.ppt

    1. 1. Robot and Intelligence Yasuo Kuniyoshi The Department of Mechano-Informatics ( Mechanics B ) & Graduate School of Information Science and Technology, Mechano-Informatics http://www.isi.imi.i.u-tokyo.ac.jp/
    2. 2. Information Foundation Information Society Information Machinery Word Computer Printing Network Multimedia Personal Computer Communication Internet Database Simulation Virtual Reality Search Engine Mobile Ubiquitous Artificial Inteligence Life Brain Cyborg Community Medical System Net Society Information Explosion e-administration Electric Money Information Economics Globalization Environment Industry Robot Body Expression Physicality Contents Media Art Entertainment Movie Music New Art “ Otaku” Culture Museum Intellectual Property Journalism Picture Cognition Information Changes the World   - the Global View - Ikuo Takeuchi “ Interface of Information and Art” Information Culture Osamu Sudo “ Information Explosion and Creation of New Network Society” Tomomasa Sato “ Robot, Information and Life” Knowledge of Information Hiroshi Harashima “ Why Information Technology Now?” Hiroshi Komiyama “ The University and Information”  - Information Changes Learning - Paper Telecommunication Information Gap Security Public Media Information Distribution Science Calculation Anime Game Drama Public Art Education Manga Film WEB2.0 Intelligent Robot Android Digital Human Mass Media Environmental Robot Neural Network Industrial production Ultimate Frontier Robot Life-supporting Robot Cell Production-supporting Robot Industrial Robot
    3. 3. Information, Robot and Life   December 13 Information Science of Robot Robot Informatics to Understand Human Cognition              ~ Yasuo Kuniyoshi ~ Robot and Intelligence ( Information Science of Robot that Learns and Develops) decision ・ judgement ・ interpretation ・ deduction ・ dialogue prediction ・ goal behavior sense activity programmed behavior code signal cognition ・ concept language assessment ・ emotion learning memory body evolution, cognition, systems cognitive development robot  ⇔  human cognitive science reflex information mechanics simulation virtual reality artificial intelligence brain nerve life robot cyborg physicality environmental robot scientific calculation ultimate frontier robot humanoid cognition life-support robot digital human cell production supporting robot industrial robot intelligent robot
    4. 4. Roots of information = physics <ul><li>Entropy : The Second Law of Thermodynamics . Quantity of state to show complexity in the air . (Clausius) </li></ul><ul><li>= Amount of information measured by observing physical state ( Szillard). </li></ul><ul><li>Amount of information for probability event (Shannon) </li></ul>The two-dimensional work of art depicted in this image is in the public domain in the United States and in those countries with a copyright term of life of the author plus 100 years. This photograph of the work is also in the public domain in the United States (see Bridgeman Art Library v. Corel Corp. ). http://en.wikipedia.org/wiki/Rudolf_Clausius Rudolf Clausius (1822-1888) Leo Szilard, near Oxford, spring 1936. Photo copyright U.C. Regents; used by permission. Contact Mandeville Special Collections Library, U.C. San Diego , for information on obtaining Szilard images. http://www.dannen.com/szilard.html Szilard, L. (1929) &quot;Über die Entropieverminderung in einem Thermodynamischen System bei Eingriffen Intelligenter Wesen&quot;, Zeitschrift für Physik 53 :840–856 Claude Shannon (1916-2001) http://en.wikipedia.org/wiki/Claude_Shannon This is a copyrighted image that has been released by a company or organization to promote their work or product in the media , such as advertising material or a promotional photo in a press kit. The copyright for it is most likely owned by the company who created the promotional item or the artist who produced the item in question ; you must provide evidence of such ownership. Lack of such evidence is grounds for deletion as established by these terms of use . It is believed that the use of some images of promotional material to illustrate: the person(s), product, event, or subject in question where the image is unrepeatable , i.e. a free image could not be created to replace it on the English-language Wikipedia , hosted on servers in the United States by the non-profit Wikimedia Foundation , qualifies as fair use under United States copyright law . Any other usage of this image, on Wikipedia or elsewhere, might be copyright infringement . See Wikipedia:Non-free content and Wikipedia:Publicity photos . Additionally, the copyright holder may have granted permission for use in works such as Wikipedia. However, if they have, this permission likely does not fall under a free license. Claude Shannon, &quot;A Mathematical Theory of Communication&quot;, Bell System Technical Journal , vol. 27, pp. 379–423 and 623–656, 1948 Boltzmann constant
    5. 5. Information  ≠  Amount of Information <ul><li>Shannon defined “amount of information”, but he did not define “what information is”. So, as a consequence … </li></ul><ul><li>Basic concept ( of pure information science ) : communication :” How can information be sent to receivers without losses” </li></ul><ul><li>Amount of sent information and amount of received information </li></ul><ul><li>Is this OK ?! </li></ul><ul><li>To know the meanings of “information”, we must think how they are generated and what they would cause (or how they are used). </li></ul><ul><li>The real phenomena : object-> object‘ : real phenomena </li></ul><ul><li>Aim for re-fusion with physices </li></ul>
    6. 6. Intelligent Robots in the Past <ul><li>They understood languages : recognized sounds , made sounds, synthesized voices, concept dictionary ・・・ static equivalency . They didn’t understand “meanings”. </li></ul><ul><li>Cognition ・ judgment ・ activity : Humans previously gave rules to robots. </li></ul><ul><li>Learning : Robots only matched given behavior elements to fit criteria for evaluation which human had determined. They cannot deal with situations which were not envisioned previously. </li></ul><ul><li>⇒” Demonstration” within the scope of the assumption was possible, but when something unexpected occurred, robot reacted absurdly. </li></ul><ul><li>Today, this is treated by broadening assumed range. </li></ul><ul><li>Symbol grounding problem : interpretation, utility </li></ul>
    7. 7. Simon’s Ant on the Beach Slides by Rolf Pfeifer (from AILectures from Tokyo) Herbert A. Simon Goal x Env x Body -> Actual motion
    8. 8. Information is generated from interactions between subjects and environment. <ul><li>Information is generated from interactions between nerve system⇔body⇔environment. </li></ul><ul><li>Physicality : constrains these interactions (meaningfully) and structuralizes them. </li></ul>sensor motor body acting body environment information system Environment dynamics learning emotion consciousness adaptation self perception Minoru Asada, Yasuo Kuniyoshi “Robot Intelligence” , Iwanami Lectures on Robotics , Iwanami Shoten , 2006.(Chapter 1 )
    9. 9. The „Didabot“ experiment arena with Styrofoam cubes Experiment by René te Boekhorst and Marinus Maris “ Didabot”: simple robot for didactical purposes Slides by Rolf Pfeifer (from AILectures from Tokyo)
    10. 10. „ Didabot“ experiment – overview Slides by Rolf Pfeifer (from AILectures from Tokyo)
    11. 11. What are the robots doing? Slides by Rolf Pfeifer (from AILectures from Tokyo)
    12. 12. Didabots s R r R s L -g LR r L + + c -g RL s R r R s L -g LR r L + + c -g RL s R r R s L -g LR r L + + c -g RL (a) (b) (c)
    13. 13. Didabots – real ants styrofoam cubes dead ants Slides by Rolf Pfeifer (from AILectures from Tokyo)
    14. 14. How brain has evolved : evolutionary tree craniata ( lamprey ・・・) eukaryote eubacteria archaea cnidaria ( hydra , jellyfish , coral , sea anemone ,・・・) porifera ( sponge ) annelida (lobworm , leech , earthworm ,・・・) deuterostome animal bacteria protoctista primordial organism mollusk ( shellfish , squid , octopus ,・・・) echinoderm ( asterid , sea cucumber ,・・・) chordata arthropod insect vertebrate fish amphibians ( frog , newt ,・・・) reptiles birds prokaryote plants mammals crustacea cephalochordata hemichordata urochordata ( ascidiacea ) gnathostamata artiodactyla ・ whale ( cow , whale ,・・・) primates ( monkey ・・・ human ) rodents ( mouse ,・・・) ・・・ ・・・ proboscidean ( elephant ,・・・) Minoru Asada , Yasuo Kuniyoshi “Robot Intelligence”, Iwanami Lectures on Robotics, Iwanami Shoten, 2006.(Chapter 1 )
    15. 15. Minoru Asada , Yasuo Kuniyoshi “Robot Intelligence”, Iwanami Lectures on Robotics, Iwanami Shoten, 2006.(Chapter 1 ) ヒドラ 細胞体 (散在神経系) (かご形神経系) プラナリア 神経節 (はしご形神経系) バッタ 神経節 脳 カエル (管状神経系) 脊髄 脳 カイメン Evolutionary path to humans single-cell organism, primordial sensorimordial system and sociality multicell stimuli reactivity Acquisition of muscle and nerve primordial reflex formation of head formation of sarcomere Sensorimordial integration Spinal cord , central nerve system, complex instinctive behavior . hierarchical regulation . Land animals, mammals, cerebral cortex, memory, expressing ability . Primates appear. Cognition of complex situation, judgements Apes appear. Analogy, deduction, social intelligence Humans appear Abstract thinking, language, education, culture Examples of existing creatures E.coli , euglena , slime mold sponge hydra , jellyfish earthworm , sea hare , squid , octopus ,( insects ) fish , frog , crocodile mouse , cat , pig ,( bird ) monkey gorilla , chimpanzee human ( early people , modern people ) Autonomous behavior, cognitive function taxis . reaction to chemicals, temperature, light, approach to nutrients, avoidance of toxin, preference for light Origin of reflex, activities of cilia on whole body stops in response to stimuli, stop in water flow . Primitive reflex, capture prey that touched tentacles, swim . reflex , releasing mechanism and programmed behavior, nerve controlled body movements . Insects have ability to adapt to environment, memorize, and learn . Various instinctive behavior : mating , nesting , territory, warming eggs etc . Behavior based on memory, map cognition, sophisticated memory, learning ability Sophisticated spatial perception and motor control for arboreal life, integration of information . Making and using tools, limited mimicry, self cognition,guessing others’ mind . Erect bipedalism, mimicry, abstract thinking, language, creativity, science, technology, art sociality Communication by transmitter substance, population synchrony,aggregation none none Genetically-controlled social insects, programmed communication between individuals, pheromone etc. . Control by instinctive behavior is predominant. . Instinctive behavior is predominant. Mating, nurturing, troops, group hunting, orders . Recognition of social relationships, identification of individuals,social hierarchy, glooming . Communication by eye contact, dynamic partnerships hierarchy . education , culture , social system , economy Basic structure of body, ecologic traits Single-cell, prokaryote/eukaryote, nutrient intake from cell membrane . Undifferentiated multi-cell, amorphous, many pores on body wall, sessile coelenteron ( eat and excrete by mouth ) Digestive tract, distinction between mouth and anus, head part can be clearly identified. Spinal cord, inner skeletal structure Terrestrial life, development of 4 limbs, viviparity, homeothermism, many are nocturnally active Hands and fingers that can grasp things (Thumbs are placed opposite ). Face . Many live on trees . Enlargement of body, omnivorous (fruits, plants, insects, animals) Curving spinal cord adopted to erection. motion mechanism, sensory organs Flagellum motor, cilia , vermiculation, sensor such as chemical/light receptor none . Nutrient intake by flagellum in a pore, generation of water flow to take up nutrients Plain muscle, no skeleton . Striated muscle (rapid),arthropods have outer skeleton link system. Sensory organs are centered on head (eyes, antennas) , sensory organ to extract specific information . 4 limbs. Inner skeleton is surrounded by muscles . Dynamic coordination is needed to keep balance. Eyeballs are developed. Sophisticated alterness ( activity improved by homeothermism ), keen senses (especially, olfactory and auditory senses) Development of fingers, Sophisticated eyesight Cleverness of fingers, various vocality . Development of sound producing organ (language sounds) . Development of facial muscles . nerve system None, some times chemical signal transmission from receptor to moving organ . none . Intercellular transmission by membrane potential in glass sponge Diffuse nervous system, no nerve center Ganglion, predominance of head ganglion, more than 100 thousand neurons Myelin sheath, faster transmission, brain-spinal cord system is established, CPG , brain stem , cerebellum, limbic cortex Development of cerebellum and cerebrum, development of cerebral cortex,formation of sensory field and motion field, development of regions related to auditory and olfactory Vision-related region in cerebral neocortex enlarged largely (cortex 55%) . Formation of association area . Large development of frontal cortex, enlargement of whole brain . Prefrontal area, joints of vertex, temporal area and occipitis TPO , large development in language area .
    16. 16. Evolution of Intelligence : information structure that environment×body create (possibility )⇔ nerve system 無顎動物(ヤツメウナギ・・・) 真核生物 真正細菌 古細菌 刺胞動物(ヒドラ,クラゲ,サンゴ,イソギンチャク,・・・) 海綿動物(カイメン) 環形動物(ゴカイ,ヒル,ミミズ,・・・) 新口動物 動物 菌類 原生生物 原初の生物 軟体動物(カイ,イカ,タコ,・・・) 棘皮動物(ヒトデ,ナマコ,・・・) 脊索動物 節足動物 昆虫 脊椎動物 魚 両生(カエル,イモリ,・・・) 爬虫 鳥 原核生物 植物 哺乳 甲殻類 頭索動物 半索動物 尾索動物(ホヤ) 有顎動物 偶蹄・鯨(牛,鯨,・・・) 霊長類(サル・・・ヒト) げっ歯類(ネズミ,・・・) ・・・ ・・・ 長鼻類(ゾウ,・・・) Primordial sensorimonitor system ( taxis ) First Communication ( intercellular signal transduction ) First reflex ( muscle and nerve ) Sensorimonitor integration and programmed behavior pattern ( head formation , central nerve system ) Hierarchical behavior control ( spinal cord and brain ) Choice of behavior, learning ( land environment, cerebrum ) Sociality based on programmed behavior ( social evolution ) Temporal-spatial memory and adaptive behavior ( cerebral neocortex ) Sociality by individual recognition, behavior recognition (cerebral neocortex, nurturing ) Tools , mimicry , understanding self and others, language, culture ・・・ 浅田稔,國吉康夫:ロボットインテリジェンス,岩波講座ロボット学,岩波書店, 2006.(1 章) Minoru Asada , Yasuo Kuniyoshi “Robot Intelligence”, Iwanami Lectures on Robotics, Iwanami Shoten, 2006.(Chapter 1 )
    17. 17. Body makes brain ! Creation and Development of Cognitive Structure Body Environment Brain Early Development Plasticity & Self-Organization Information-Driven Esp. Cerebral Cortex Emergent Information Structure from Embodied Interaction
    18. 18. Creative/Developmental Structuring Theory <ul><li>  Find out the most basic principle of generation and changes of intelligent behaviors, not by making a complete model of an intelligent behavior, but by removing unimportant things from it. Construct this in a real environment , monitor how it behaves and develops, and scale it up by feeding back to the principle. </li></ul><ul><li>Basic principles of intelligence creation and development : </li></ul><ul><ul><li>physicality = information structure formed by a body </li></ul></ul><ul><ul><li>A system to discover, obtain and use that information structure </li></ul></ul>Minoru Asada , Yasuo Kuniyoshi “Robot Intelligence”, Iwanami Lectures on Robotics, Iwanami Shoten, 2006.(Chapter 1 ) Understand the mechanism to create intelligence behavior Generating principles creative structuring theory assessment, choice purpose creation deduction, construction body,environment adjustment developmental structuring theory
    19. 19. “ Wisdom” of a Body With Brain Completely Removed
    20. 20. Jumping and Landing : Physical Intelligence <ul><li>Jumping and landing </li></ul><ul><ul><li>Extremely fast & dynamic – Feedback control is difficult. </li></ul></ul><ul><ul><li>Interactions with ground – Modeling and prediction are difficult. </li></ul></ul><ul><ul><li>Role of body dynamics is important. Utilization is essential. </li></ul></ul><ul><li>Wisdom : With factors above, jump and land stably. </li></ul>Niiyama ・ Kuniyoshi 05-06
    21. 21. Bio-mechanism of Legs <ul><li>Body structure ⇒natural movements </li></ul><ul><ul><li>Bi-Articular Muscles </li></ul></ul><ul><ul><li>McKibben pneumatic actuators </li></ul></ul><ul><ul><li>Size of each parts ・ distribution of mass </li></ul></ul>Bi-Articular Muscle Mono-Articular Muscle Ham Strings Bi-Articular Muscle Niiyama ・ Kuniyoshi 05-06
    22. 22. Jumping & Landing: MOWGLI Niiyama & Kuniyoshi 05-06 Ryoma Niiyama , Yasuo Kuniyoshi : Development of Jumping and Landing Robot Based on Musculoskeletal Bio-Mechanics , the 11 th Robotics Symposium , 1C1 , pp.50-55, Saga , March , 2006 .
    23. 23. “ Wisdom” of Body <ul><li>Adoption to Disturbance </li></ul>muscle ( including bi-articular muscle ) Articular motor ( not including bi-articular muscle ) Movements without disturbance ( muscle = motor )
    24. 24. Knack and Eye for Good - information structure of physicality - What is information structure shared by objects with human-like bodies ? Information structure created from interactions between body and environment Information structure controlled by brain Information transmitted by mimicry or teaching “ Once you get a knack, a work can be done very easily and securely .” “ Person with a good memory knows a good thing when he/she sees it.” -> Anyone thinks so, but this is the most important phenomenon related to the principle of human intelligence.
    25. 25. “ Roll-and-Rise” Motion -- An example emphasizing ”knacks” <ul><li>Body dynamics is activated. A skill is needed. </li></ul><ul><li>Perfect modeling is impossible ( minor drive , strong disturbance ) </li></ul><ul><li>Control switching is needed ( multiple dynamics ) </li></ul><ul><li>Not limit cycle . Includes divergence . Target aspiring. </li></ul>Yasuo Kuniyoshi , Akihiko Nagakubo , Yoshiyuki Omura , Kohshi Terada , Tomoyuki Yamamoto , Shinichiro Nagatoku   1996--2005
    26. 26. Non-Uniform Trajectory Bundle <ul><li>There are regions where phase-space ( knee-waist articular space) orbit converge or diverge. </li></ul>Standing position ( convergence ) Lifting and lowering legs ( divergence ) Sole landing ( convergence ) T. Yamamoto and Y. Kuniyoshi 02
    27. 27. Non-uniform Converging Point K. Terada and Y. Kuniyoshi 04 entropy Amount of information
    28. 28. Global Dynamics -> Good Point ( physical information structure ) <ul><li>“ Good point”    ←“ knack” </li></ul><ul><ul><li>Centered point of information </li></ul></ul><ul><ul><li>Stable to disturbance </li></ul></ul><ul><ul><li>Compatibility of adaptation and assurance </li></ul></ul><ul><ul><li>Creation from body, environment, purpose </li></ul></ul><ul><ul><li>Symbolic information shared by “human-type”. </li></ul></ul>
    29. 29. Informatics of Dynamics Lyapunov exponentλ Predictability:How long can information at the initial state be retained. Skillful motions : Achieves another knack during procedure:Movements grasping information ->  -> Kolmogorov-Sinai entropy Distance between orbits.)
    30. 30. Success! – Rising in 2secs. Gyro-based balance control invoked at the instant of feet landing Ohmura, Terada & Kuniyoshi 03 Yasuo Kuniyoshi, Yoshiyuki Ohmura, Koji Terada, Akihiko Nagakubo: Dynamic Roll-And-Rise Motion By An Adult-Size Humanoid Robot, International Journal of Humanoid Robotics, vol.1, no.3, pp.497-516, 2004.
    31. 31. Tactile Sensor 180 mm 110 mm Small tactile sensor Cut and paste implementation Light diffusing method Ohmura , Nagakubo , Seta , Kuniyoshi 2005 Yoshiyuki Ohmura, Yasuo Kuniyoshi, Akihiko Nagakubo:Conformable and Scalable Tactile Sensor Skin for Curved Surfaces, Proc. IEEE Int. Conf. on Robotics and Automation, pp.1348-1353, 2006.
    32. 32. Implementation to Humanoid Soft touch skin back ( 308 points ) Soft touch skin for body (192 points) hip ( 150 points ) sole 56 points each 400mm 400mm groin ( 62 points ) Ohmura, Kuniyoshi 2006 (stepping stone to close contact with humans … )
    33. 33. Rising Motion Experiment Ohmura, Kuniyoshi 2006 Yoshiyuki Ohmura , Yasuo Kuniyoshi : Physical rising motion of a humanoid using distributed tactile senses , the Robotics Society of Japan, the 24 th Academic Lecture , CD-ROM , 2H22, 2006.
    34. 34. Temporal focuses in understanding actions Yasuo Kuniyoshi, Yoshiyuki Ohmura, Koji Terada, Akihiko Nagakubo, Shin'ichiro Eitoku, Tomoyuki Yamamoto: Embodied Basis of Invariant Features in Execution and Perception of Whole Body Dynamic Actions --- Knacks and Focuses of Roll-and-Rise Motion, Robotics and Autonomous Systems, vol.48, no.4, pp.189-201, 2004.
    35. 35. When do you know it’s action X? Temporal localization of action information
    36. 36. When do you know it’s action “X”? Temporal localization of action information <ul><li>30S’s ( M23,F7) </li></ul><ul><li>64 trials </li></ul><ul><li>=2 performers x (2 succ.+ 2fail.) x 8 samples (diff. Length). </li></ul><ul><li>Random display </li></ul><ul><li>Guess succ./fail. </li></ul>Eitoku&Kuniyoshi 04
    37. 37. Flow-in of information⇒Percentage of correct answers increases. ( entropy decrease )
    38. 38. First Summary <ul><li>Physical interactions between body and environment generate information. </li></ul><ul><li>Why can humans understand each other’s behavior? Why can we communicate with each other by words? A search for basic principles leads to the question of why behaviors and concepts are common to every humans. Now, it is being understood that similarity of bodies have a key role in this commonness. </li></ul>
    39. 39. Body Makes Brain <ul><li>Neuron projection at the early stage of development is dependent to activities. [Crair 1999] </li></ul><ul><li>Column structure of cerebral cortex can be explained by self-organizing model. </li></ul><ul><li>Plasticity of various inner brain expressions such as in somatosensory area </li></ul><ul><li>Environmental factors and gene expression interact with each other through regulatory gene network . There is no one-sided control by gene program.[Ridley 2004] . Behavior and learning of an individual take part in this interactions, and consequentially, they control development. </li></ul><ul><li>FOXP2 which became known as a “language gene” is speculated to be a gene for controlling activities, which also affects language development.[Johnson 2005] . </li></ul><ul><li>There are studies that point out retarded development of motion control might be a cause for autism. </li></ul>
    40. 40. Plasticity of Somatic Sense Map <ul><li>Monkeys : The map changes by amputation of a finger or a nerve : adjacent regions penetrate in. </li></ul><ul><li>By suturing 2 fingers, boundary disappears. </li></ul><ul><li>Humans : 10 days after separating congenital adhesion, map of a finger changes. ( brain inducing map ) </li></ul><ul><li>Monkeys : After a training to use tip of fingers, region widens ( right ) </li></ul><ul><li>String musicians have larger regions for left fingers. </li></ul>Kandel, et al.: Principles of Neural Science, McGraw-Hill, 2000.
    41. 41. Hypothesis <ul><li>Structures of body-environment interactions guide self-organization of nerve systems, and are deeply related to construction of cognitive mechanisms. </li></ul><ul><ul><li>General structures of body and nerve system are controlled by genes, and cognitive functions are controlled by information-driven self-organization . </li></ul></ul><ul><ul><li>By detecting such structures and exposing them, mechanisms reflected to learning and self-organization are studied. </li></ul></ul><ul><li>Development of motor function is a basis for all cognitive functions. </li></ul><ul><ul><li>Coherent learning information generate from movements. </li></ul></ul><ul><ul><li>Senses accompanied to motions have ultra modal consistency. </li></ul></ul>
    42. 42. Perception, Behavior and Learning During Embryo Period <ul><li>Generalized movement: </li></ul><ul><ul><li>Starts from 2 months after impregnation.(Kisilevsky&Low98,Joseph00) </li></ul></ul><ul><li>Vision : </li></ul><ul><ul><li>Eye opening : 20 weeks after impregnation (Lecanuet&Schaal96), 40% time rate 34 weeks after impregnation (Birch&O’Conner01). </li></ul></ul><ul><ul><li>light : 10 % of outer light (red) reaches womb ( by animals , Jacques et al.87) </li></ul></ul><ul><ul><li>retina : All cell layers are formed 7 months after impregnation , middle peripheral vision starts its function 30 weeks after impregnation. </li></ul></ul><ul><ul><li>optic nerve : forms before 28 weeks . </li></ul></ul><ul><ul><li>cognition : 32 weeks ( 8months ) immature baby preferential looking </li></ul></ul><ul><li>Adoption ( learning ?): </li></ul><ul><ul><li>Naturalization and denaturalization : observation by body movements reaction of embryo to sound signal (Madison et al 86) . </li></ul></ul><ul><ul><li>Stimuli combinations during embryo period is effective after birth (Lecanuet&Schaal96). </li></ul></ul><ul><li>Mutual bonding of nerve circuit network : </li></ul><ul><ul><li>There are many random bondages in an embryo brain. There is a peak little before birth . (Rakic et al.86) </li></ul></ul>
    43. 43. Development of Embryo Movements Sangawa & Kuniyoshi 06
    44. 44. Two-Factor Model of Motor Development <ul><li>Spontaneous Exploration of Motor Patterns </li></ul><ul><ul><li>Emergence of body-afforded movement patterns </li></ul></ul><ul><ul><li>Driven by chaos-generating Medulla-Spine (Bulbospinal) system ( Small ‘ 99 ) . </li></ul></ul><ul><li>Learning of Sensory-Motor Patterns </li></ul><ul><ul><li>Re-usable units from emergent patterns. </li></ul></ul><ul><ul><ul><li>Controllable units : sensory-motor correlation. </li></ul></ul></ul><ul><ul><ul><li>Learning of signaling patterns. </li></ul></ul></ul><ul><ul><li>Self organization of motor-somatosensory areas. </li></ul></ul>Sangawa & Kuniyoshi 06
    45. 45. ・ Body Models of Embryo and Neonate <ul><li>Musculoskeletal system (198 muscles) </li></ul><ul><li>Muscle kinetics model (He et al. 01, Hill 38) </li></ul><ul><li>Size, mass, inertia are set based on papers </li></ul><ul><li>Movable limit of joints , setting of natural position </li></ul><ul><li>Growth!? (parameter: gestational age). </li></ul>Sangawa & Kuniyoshi 05 Touch sensor: 1448 points
    46. 46. Partial Model of Central Nerve System The central nervous system figure above is adopted from: Eric R. Kandel (Editor), James H. Schwartz (Editor), Thomas M. Jessell : Principles of Neural Science, McGraw-Hill , 2000, ISBN: 0838577016. (Hard/pap) Sangawa & Kuniyoshi 06
    47. 47. Sensory Organ Models <ul><li>Spindles (Ia output) </li></ul><ul><ul><li>Linear increase wrt. muscle length (static condition). </li></ul></ul><ul><ul><li>Sensitive to muscle stretch speed. </li></ul></ul><ul><ul><li>Sensitivity modulation by α,γinputs. </li></ul></ul><ul><li>Golgi tendon organ (Ib output) </li></ul><ul><ul><li>Approx. proportional to muscle tension. </li></ul></ul>Sangawa & Kuniyoshi 06 based on He et al. 01
    48. 48. Spine Model <ul><li>Stretch Reflex ( Spindle->Ia->α->Muscle->Spindle ) </li></ul><ul><ul><li>Regulate muscle length. </li></ul></ul><ul><ul><li>Postural control. </li></ul></ul><ul><li>Ib inhibition ( Tendon->Ib->α->Muscle->Tendon ) </li></ul><ul><ul><li>Regulate muscle tension. </li></ul></ul><ul><li>α-γlinkage </li></ul><ul><ul><li>Simultaneous activation of α and γ. </li></ul></ul><ul><ul><li>Override feedback loop of stretch reflex. </li></ul></ul><ul><ul><li>More force on contracted muscles. </li></ul></ul><ul><ul><li>-> Voluntary motion. </li></ul></ul>Sangawa & Kuniyoshi 06 Spinal neurons transfer function (He et al. 01):
    49. 49. Medulla Model <ul><li>Each neuron controls 1 muscle. </li></ul><ul><li>No direct coupling between CPG's </li></ul><ul><li>CPG coupling through the body. </li></ul><ul><li>Periodic/chaotic movement depending on inputs (constant + M1 signal). </li></ul>Sangawa & Kuniyoshi 06
    50. 50. Chaotic motor exploration <ul><li>Coupled CPG's generate periodic / chaotic signals (Asai 03). </li></ul><ul><ul><li>Periodic for uniform control inputs. </li></ul></ul><ul><ul><li>Chaotic for non-uniform control inputs. </li></ul></ul><ul><li>Motor exploration by couple chaotic system ( Kuniyoshi&Suzuki 04 ) </li></ul><ul><ul><li>Coupling of chaotic elements via embodiment. </li></ul></ul>1CPG 1CPG Sangawa & Kuniyoshi 06 BVP eq.
    51. 51. Dynamics of S1 and M1 <ul><ul><li>三田ステーションビル「アミタ」 </li></ul></ul>Active potential of S1 neuron i Input from S0 neuron j to S1 neuron i Input from adjacent S1 neuron k Active potential of M1 neuron i Output of S1 neuron j corresponding to M1 neuron i Output of adjacent neuron to S1 neuron j S1 M1
    52. 52. <ul><li>input : </li></ul><ul><ul><li>synapse->dendrite </li></ul></ul><ul><ul><li>synaptic efficacies, weight </li></ul></ul><ul><li>integration : </li></ul><ul><ul><li>total </li></ul></ul><ul><ul><li>cell body </li></ul></ul><ul><ul><li>membrane potential </li></ul></ul><ul><li>output : </li></ul><ul><ul><li>threshold </li></ul></ul><ul><ul><li>axon </li></ul></ul><ul><ul><li>action potential </li></ul></ul>Basic Activities of Neuron 塚原仲晃編,脳の情報処理,朝倉書店, 1984.
    53. 53. Neuron Model Output value of neuron j Bias node for neuron i Bonding load from neuron j to i (synapse load) Inner potential of neuron i Linear or Simple threshold function or Sigmoid function 0 x i x j ・ ・ ・ f ∑ w i0 x 0= 1 × ・ ・ ・ × w ij
    54. 54. Neuron Model and Learning <ul><li>Hebb learning : When 2 neurons are activated at the same time, bond between them is enhanced. </li></ul><ul><li>Correlated learning </li></ul>Learning efficiency x i x j ・ ・ ・ f ∑ w i0 x 0= 1 × ・ ・ ・ × w ij
    55. 55. Lateral Inhibition <ul><li>Mexican-hat function </li></ul><ul><li>Promotion of clusters ( gather similar things ) </li></ul>When combined with Hebb learning, self-organization learning. Position of neurons Strength of excitement Strength of inhibition Excitement radius Inhibition radius x 1 x j y 1 … y i w ij
    56. 56. Character Map of Cerebral Primary Visual Area Bonhoeffer and Grinvald, 1991
    57. 57. Motor ( M1 ) -SomatoSensory ( S1 ) Areas Model <ul><li>Self-Organizing Network with spatial structure ( Chen 97 ) </li></ul><ul><li>Somatosensory area ( S1 ) </li></ul><ul><ul><li>Competitive SOM </li></ul></ul><ul><ul><li>S0-S1: Modified Hebbian, fully connected. </li></ul></ul><ul><li>Motor area ( M1 ) </li></ul><ul><ul><li>Receptive field around corresponding point on S1 </li></ul></ul><ul><ul><li>M1-α , M1-γ , M1-CPG Modified Hebbian </li></ul></ul>Sangawa & Kuniyoshi 06
    58. 58. Left and Right Hemisphere <ul><li>Left and right neural system </li></ul><ul><ul><li>S1, M1:20x10 each neuron </li></ul></ul><ul><ul><li>Spinal cord, medulla : 99 neurons </li></ul></ul><ul><ul><li>Bridge : link left and right S1,M1 </li></ul></ul><ul><ul><li>All bonds ( excitement ) </li></ul></ul><ul><ul><li>Hebbian rule ( altered version ) </li></ul></ul><ul><li>Restriction input </li></ul><ul><ul><li>Constant value ( 0.5 ) </li></ul></ul><ul><ul><li>Smooth motion </li></ul></ul>Sangawa & Kuniyoshi 06 Shinji Samukawa, Yasuo Kuniyoshi : Body and Brain-Spinal Cord Model of Embryo and Neonate, Self-Organization of Somatic Sensory Area and Motor Area , the Robot Society of Japan, the 24th Academic Lecture , CD-ROM , 2L24, 2006.
    59. 59. Embryo and Neonate Environment <ul><li>embryo ( 35 weeks after impregnation) </li></ul><ul><ul><li>Womb environment : gravity, buoyancy, fluid resistance, navel string (holds a body at a point of belly button) </li></ul></ul><ul><ul><li>Uterine wall : non-linear spring, damper model </li></ul></ul><ul><li>neonate ( 0 week after birth ) </li></ul><ul><ul><li>Normally, gravity </li></ul></ul><ul><ul><li>Flat floor </li></ul></ul><ul><ul><li>Surrounded by a fence </li></ul></ul>Sangawa & Kuniyoshi 06 寒川新司,國吉康夫:胎児・新生児の身体・脳脊髄モデルと体性感覚野・運動野 の自己組織化,第24回日本ロボット学会学術講演会, CD-ROM , 2L24, 2006. Shinji Samukawa, Yasuo Kuniyoshi : Body and Brain-Spinal Cord Model of Embryo and Neonate, Self-Organization of Somatic Sensory Area and Motor Area , the Robot Society of Japan, the 24th Academic Lecture , CD-ROM , 2L24, 2006.
    60. 60. Movements of Embryo Model Sangawa & Kuniyoshi 06
    61. 61. Movements of Embryo Model Sangawa & Kuniyoshi 06
    62. 62. Pattern of Cortex (under consideration) <ul><li>embryo model </li></ul><ul><ul><li>S1  :  classify roughly to legs, head and body . </li></ul></ul><ul><ul><li>M1->α , γ  :  Differentiation corresponding to each parts of body </li></ul></ul><ul><li>newly-born model ( no womb experience ) </li></ul><ul><ul><li>S1  :  Head and neck regions are predominant. </li></ul></ul><ul><ul><li>M1->α , γ  :  Undifferentiated compared to embryo model. </li></ul></ul>
    63. 63. Interpretation(under consideration ) <ul><li>embryo </li></ul><ul><ul><li>Motion restriction by uterine wall </li></ul></ul><ul><ul><ul><li>Rotating movements of legs and neck : generation of cooperative relationship </li></ul></ul></ul><ul><ul><ul><li>S1 differentiation drives M1 differentiation . </li></ul></ul></ul><ul><li>neonate (no embryo experience) </li></ul><ul><ul><li>Non-restricted motions </li></ul></ul><ul><ul><ul><li>Each part moves independently ( no cooperation ) </li></ul></ul></ul><ul><ul><ul><li>Neck is predominant ( Muscle of neck is strong ) </li></ul></ul></ul><ul><ul><li>M1 is undifferentiated. </li></ul></ul><ul><li>Motion restriction by uterine wall might play a great role in early development of motions. -> for cognition development .. </li></ul>
    64. 64. Summary and Issues <ul><li>Undeveloped neural systems move human-type body </li></ul><ul><li>-> These systems detect various motions autonomously and repeat those motions. </li></ul><ul><li>-> Cerebral cortex reflects its structure and self-organize. </li></ul><ul><li>Information structure given from human-like body makes brain. : Body diagram, picture of a body, behavior concept, object concept . </li></ul><ul><li>Can humans understand each other, using shared structures? </li></ul><ul><li>Would a brain developed like this possess human mind? </li></ul>
    65. 65. Steps from Motion to Cognition <ul><li>GM->structuralizing by body and environment dynamics->motion units </li></ul><ul><li>Body diagram , integration with view, body scheme </li></ul><ul><li>Prediction and segmentation of motions accompanied sense , attention , consciousness </li></ul><ul><li>Attention to other people’s motions and turn taking </li></ul><ul><li>Others’ body diagram can be obtained by body diagram. </li></ul><ul><li>Motivation of identifying others, mimicry trial </li></ul><ul><li>Stand holding on to something </li></ul><ul><li>Acquisition of language </li></ul>
    66. 66. Conclusion <ul><li>Information structure generated from interactions between nerve, body and environment . </li></ul><ul><li>Neural system to drive and learn it . </li></ul><ul><li>Self-organized information in brain communicate with others, interpret information and decide what to do. . </li></ul><ul><li>By imagining developing robot, all procedures through generation of information to interpretation and uses can be considered as a closed system. </li></ul><ul><li>For a robot that can really communicate with humans .. </li></ul>

    ×