KAWATO Mitsuo
Computational Neuroscience Laboratories


Introduction
　　ATR opened up the field of Computational Neurosciences, the approach of which is expressed by the catch phrase "Knowing the brain by creating brains," and has become a world leader in this field. The goal of the Humanoid Brain Project is to further develop these research methods, and to give humanoid robots artificial brains that are based on neuroscience. The technologies that evolve from the ripple effects of this process will give rise to applications in rehabilitation, household robots, and the creation of futuristic communication terminals.

Advancements in human communication technologies
　　It would not be an exaggeration to say that the developments in human culture, science and technology, industry, and economics are entirely dependent on the new communication skills and technologies invented by mankind. Three million years ago, our human ancestors had already acquired communication skills called mimesis, which separated them from anthropoid apes, and had become capable of manufacturing tools and working in groups. With the development of language some tens or hundreds of thousands of years ago, the efficiency of communication increased, and the diversity of stone tools expanded at an explosive pace. As shown in Table 1, the creation of each new communication technology, such as writing, the telegraph, or the Internet, brings about great leaps in human culture and civilization, such as those reflected in the ancient agrarian civilizations, the industrial revolution, and more recently the IT revolution. If we extrapolate from past trends, there is no question that our future communication technologies will surpass the boundaries of time, space, and culture, and will exceed the limited information of the written word, images, and sounds, achieving a natural form of communication that incorporates face-to-face physical interactions.

Table 1. Future communication technologies.
3 million yrs. ago Mimesis Collaborative work
200,000 - 30,000 yrs ago Language High efficiency
Several thousand yrs. ago Codes Surpassing boundaries of time and space
Several hundred yrs. ago Transportation (physical) Not in real time
160 yrs. ago Telegraphs Codes / real time
120 yrs. ago Telephone Spoken language / emotions
20 yrs. ago Internet Codes → Passive images / audio
Future Mimesis + time + space + culture ;
Agent communications		 Brain research + robots


Difficulties to be overcome by future communication technologies
　　Let us assume that a husband and wife who enjoy playing tennis are living apart because the wife lives in Japan and the husband has been stationed in the U.S. for work. Nevertheless, the two want more than anything to be able to play tennis together. In order to actually play tennis (to experience the physical feelings), as shown in Fig. 1, there would have to be an "agent" robot of the husband near the wife, and an "agent" robot of the wife near the husband, with the two playing tennis in Japan and the U.S. at the same time. The greatest obstacle in enabling these two people to play simultaneously is to overcome the time delays that accompany communications. Anybody who has had the unpleasant experience of delays of several seconds in verbal and facial communication cues during teleconferencing using satellites would think that it is impossible to overcome these time delays. The only way to resolve this problem would be to mount a quantitatively good model of the husband's brain into the husband's agent robot, though only for a short time. We once heard a comment that in a sense, this is equivalent to creating a short-term time machine.


Figure 1. The husband's brain in the husband's agent robot.

Integrating robot research and brain research
　　As explained above, if we can successfully develop a more human-like humanoid robot based on an understanding of the basic functions of the brain, then we will be much closer to achieving revolutionary futuristic communication terminals. The problem of time delays can only be resolved by installing into terminals quantitative brain models of the persons communicating, and by giving these terminals the ability to imitate the users. This is an indication that the field of telecommunications will eventually demand an integration of communication technologies based on both neuroscience and robotics.


Photo 1. DB showing off its skills.

DB learns by imitation
　　The humanoid robot DB was developed through the ERATO Kawato Dynamic Brain Project. Through the research conducted at the ATR Cyber Human Project, before it had reached its 3rd birthday, DB was able to perform the 24 skills shown in Table 2 and Photo 1. Most of these skills were acquired through watching and imitating, or are based on the brain's computation principles. Even now, robots and computers have yet to achieve the level of a two- or three-year-old child in terms of the abilities that an average human takes for granted, such as understanding conversations, understanding visual scenery, and moving smoothly. This is an indication that research in robotics and artificial intelligence has become bogged down, and at the same time shows that we have yet to gain an understanding of the functions of the human brain. In order to overcome these difficulties, we have been promoting research starting with a humanoid robot that has the same size, weight, flexibility, and sensory organs as a human, and having this robot watch and imitate human beings. "Watching and imitating" means observing the behavior of another person, and inferring the intent of that other person to resolve the same issues. We have advocated this as the basis of human intelligence.

Atom Plan
　　Right now, Japan is not well. Both the public and private sectors are looking to the scientific world for quick-fix measures to promote industry and the economy, but they are far off the mark. I believe that now is the time to begin projects with visions that will become the basis for Japan's development 10 years or 20 years down the road. For the past few years, we have been advocating the Atom Project, with an awareness of the Apollo plan in the United States. Currently, ATR and other organizations in Japan are the world leaders in the field of robotics research, and the Computational Neurosciences advocated by ATR have gained broad recognition as a revolutionary method in the area of Neurosciences, becoming an important topic of integrated research and a seed for a new vision for Japan. Our goal, by achieving an organic combination of these technologies, methods, and knowledge, is to gain an understanding of the functions of the brain, and a unified understanding of communication functions in particular, and then to develop a learning humanoid robot based on this understanding. We will aim to develop, within the next ten years, a completely independent humanoid robot with the intelligence of a five-year-old, and the ability to sense and to move. The Apollo Project inspired a generation disheartened by the "Sputnik Shock" by setting a goal that anyone could understand - that of sending a man to the moon. This same project left behind a vast legacy for America, ranging from Teflon pans to computer sciences. At this point, we can enjoy the wonder of not knowing whether the technologies born from the ripple effects of the Atom Plan will manifest themselves in mobile phone terminals, entertainment robots, long-term care robots, or rehabilitation scenarios. The most important thing, however, is that the goal of the project must have a vision, and that the researchers and engineers must be able to devote themselves to the challenges of that project. Lately, we have seen far too many foolish measures, in which huge sums of money are thrown away for the sake of so-called "quick-fix" effects. I hope that the political and economic world will soon come to reevaluate their approach.

Table 2.24 performances by DB.
Learning from Demonstration

(1) Okinawa Dance Imitation (Kachyaasi)
(2) Rock'n Roll Dance Imitation
(3) Pole Balancing Imitation
(4) Tennis Swing Imitation
(5) Real-Time Visual Tracking of Human Motion
(6) Punching Imitation
(7) Juggling
(8) Devil Stick
(9) Real-Time Hand Movement Imitation
(10) Air Hockey Imitation
(11) Tumbling a Box
(12) Moving a Small Box and Robota
Eye movement (13) VOR Adaptation
(14) Smooth Pursuit Learning
(15) Saccade
(16) Combination of 3 Eye Movement Primitives
Task dynamics Physical interaction Learning (17) Paddling
(18) Learning of Visuo-Motor Transformation
(19) Catching a Ball
(20) Drumming Joint-Performance
(21) Sticky Hand
(22) Non-Calibrated Visuo-Motor Transformation
(23) Yo-yo and Slinky
(24) Flexible Object Manipulation