(Translated from: Hans-Dieter Burkhard / Hans-Arthur Marsiske: Endspiel 2050 Wie Roboter Fußball spielen lernen)
QUESTION: Mr. Kitano, what was your personal way to robotics? Do you remember the first time you heard about robots?
KITANO: Since I grew up in Japan, I heard about robots quite early. When I was a kid I read cartoons like ìAstro Boyî or saw robots in movies. But I havenít been exposed to real robots before I became a researcher. When I was staying at the Carnegie Mellon University in Pittsburgh, which has a big robotics institute, I took a serious look at robotics for the first time.
QUESTION: You studied at Carnegie Mellon University?
KITANO: I was there as a visiting researcher from 1988 until 1993.
QUESTION: That was exactly the time when the idea of RoboCup emerged. Can you tell us how this idea took shape?
KITANO: I was doing research in artificial intelligence then and was especially interested in multiagent systems. But there seemed to be no specific applications of this technology that might have an impact on society. I also watched a robotics competition at a conference in 1991 with robots moving very slowly, then standing still for several minutes, moving again a little bit and so on. It was very boring and I thought it might be useful to define an interesting, challenging topic that requires fast moving robots to realize. That was one starting point for RoboCup. At the same time I was organizing a workshop ìGrand Challenge AI Problemsî which called for proposals for very big research issues and their possible applications in artificial intelligence and robotics that may have a high impact in the society during the next 10 to 20 years. Actually, RoboCup was not included in those discussions, but the general consensus was that to promote artificial intelligence research we needed some very symbolic and visible goal that is widely appealing. Later, when I thought about soccer as such a possible goal, I found out that soccer playing robots fit very well to developing basic technologies for a broad range of applications. That was around 1992/93. Then we spent two years on feasibility studies and in 1995 made an announcement on the International Joint Conference on Artificial Intelligence (IJCAI) in Montreal that there would be the first international RoboCup competition at the next IJCAI which was scheduled to take place 1997 in Nagoya, Japan.
QUESTION: Did you consider any alternatives to soccer as a testbed for cooperating autonomous robots?
KITANO: Yes, I thought about various applications. First I considered welfare or desaster rescue as possible testbeds. But it was difficult to find one single big focus. Desaster rescue for instance is very different from country to country. In Japan earthquakes are our major concern which is of little relevance to New York City. The people there are more concerned about crime. In other parts of the USA hurricanes are a major issue, while in Europe you have to deal with different types of floodings. On the other hand to build welfare robots you need very advanced technologies that might need 20 years to develop. It is very difficult to justify such a big project and come up with the necessary money. Rather than going directly into specific applications I thought it might be a better strategy to have something a little bit off the road, like a game where you can develop basic technologies and then transfer them to the real world. I thought about soccer immediately but also considered other sports like baseball or volleyball. But those games are much less dynamic than soccer. Volleyball for instance has a very clear distinction between the two teams which are separated by the net. Itís similar with American football and rugby where you also have a clear distinction between attackers and defenders. Also, itís very much about direct body tackling that can be solved very easily by building very strong robots. That is not what we want with RoboCup. We donít want robots that harm people. So at the end it was soccer that was left as the best alternative. It has very high dynamics, requires a high level of teamwork, does not aim to harm the opponent intentionally, and has a great popularity throughout the world.
QUESTION: You didnít intend right from the start to organize RoboCup as an international competition?
KITANO: No, in 1993 when these ideas emerged I originally planned to call it ìRobot J-Leagueî, because almost at the same time the Japanes professional soccer league ìJ-Leagueî had been founded. But I immediately received e-mails from Europe and the USA which proposed to do it as an international activity not only a Japanese one. So we changed the name into Robot World Cup, or RoboCup in short.
QUESTION: At that time, did you already have in mind the long-term goal of winning the soccer world championship with robots agains humans?
KITANO: In my first paper I wrote about creating a robot soccer team at the end of the 21st century that could beat the Brazilian national team. Immediately I got an e-mail from Italian researchers who said beating the Brazilians wouldnít be good enough. We would have to beat the Italians, too. But in general the community was quite excited by the perspective of winning the soccer world cup. Then I was on a tv interview with a very famous Japanese science journalist who mentioned that the target date at the end of the century might be too late and should be shifted to the middle of the century. He argued that it took only 66 years from the first motorized flights of the Wright brothers to the first humans landing on the moon. Since soccer robots wouldnít need to play on the moon but on Earth, 50 years should be enough. I thought about it and I found other interesting facts. For instance, the first commercial jet airliner went into service in 1947, less than 50 years after the Wright brothers. Also, it took around 50 years from the first computers like ENIAC to ìDeep Blueî which was the first to beat the human chess world champion. The same with the discovery of the structure of the DNA and the complete sequencing of the human genome. A time frame of 50 years seems to be quite adequate for reaching challenging scientific and technological goals.
QUESTION: What is the relation of RoboCup to other robot competitions? You once classified RoboCup as a ìlandmark projectî. Can you explain that?
KITANO: There are a lot of robot competitions throughout the world. But as far as I see, most of them are either for the hobby robot builder or for student education. There are a few competitions focusing on research, too. Among them RoboCup may be the single largest one. But competition is only one part of RoboCup. We also have scientific conferences, special interest groups, industrial technology transfer, and other projects going on. So itís not only about winning the competition. The real goal of RoboCup is to push the technology forward. That is one big difference. Also, no other competition that I know of has a long-term goal like winning the world championship in 50 years. This goal allows us to define milestones and adapt the rules to the progress of the technology, for instance by removing the walls on the side of the playing field or using a bigger ball. This focus on research and technology development makes RoboCup very different from most other robot competitions.
QUESTION: How did the idea of the different leagues develop?
KITANO: Originally we started with three soccer leagues: simulation, small size, and middle size. Simulation is necessary because with real robots it is very difficult to do the complex artificial intelligence research in advanced decision making or learning. Researchers needed the opportunity to do this research right away. So we decided to create a simulation based league that provides this opportunity. The small size league was motivated by two reasons. The first is that you need much space to prepare for participation in the middle size league. Many teams donít have that space, so the small size league offers them an opportunity to participate in RoboCup. At the same time we slightly changed the rules for the small size by allowing sensors outside the field, while in the middle size league all sensors have to be on board of the robots. This is the second motivation for the small size league: We want to investigate real time sensor fusion for fast moving objects. Actually, that can be applied to intelligent traffic systems. An autonomous vehicle might get its information from on-board sensors as well as from GPS satellites or sensors on the streets like magnetic loops, radar, or cameras. How to integrate this information is a very important research topic. It can also be applied to navigation of robots in the house, office, or other special environments. The middle size league on the other hand with fully autonomous robots explores the technology needed for the exploration of strange and unknown environments like other planets, disaster areas, or volcanoes where there is no chance to deploy a sensor system first. The robot has to act completely on its own.
QUESTION: This original system of leagues started to evolve quite quicklyÖ
KITANO: When Sony introduced the four-legged robot ìAiboî we thought that was very interesting. It is a very well done hardware. After some discussions we agreed that Sony should host the legged robot league with Aibo as standard platform. It turned out to become a very popular league because Aibo looks very cute and appeals to many people. Finally, in 2002 we introduced the humanoid league which we wanted to have right from the start for obvious reasons: If you want to play soccer against humans you need to have robots that are humanoid in shape and walk on two legs. That is how the RoboCup soccer leagues developed. Parallel to that we also started the rescue leagues in 2001, because we want to apply the technology to serious issues. Building soccer robots is not a goal in itself but only a starting point. As I told you, this was the idea of RoboCup right from the start. But to actually introduce the rescue league we had to wait for active participants like Satoshi Tadokoro from Kobe University or Robin Murphy from the University of Central Florida to take the lead on this. Actually, one month after introducing her robots at the RoboCup world championship 2001 in Seattle Robin Murphy deployed them at the remains of the World Trade Center in New York and was able to find eight human bodies. Of course, this was a very sad incident. But for robotics it was very important to proof the technology as effective in a real disaster environment. Meanwhile governments in Japan and Europe show great interest in this technology and we created an independent national research organization in Tokyo and Kobe. Finally, as a third division in RoboCup besides soccer and rescue we have RoboCup junior. This is about education. If we want to build by 2050 robots that can beat the human soccer world champion we need a constant flow of talents and young researchers. That is not only beneficial for RoboCup but for the whole community of artificial intelligence and robotics research. Because of this aim RoboCup junior is not mainly concerned with robot soccer. Instead we have robot dance, robot rescue, robot performance, and other broad programs so that the kids can be very creative. What we also emphasize in RoboCup junior is teamwork. So we use the robots as a tool for a more general education about science, technology, and society.
QUESTION: The goal of beating humans on the soccer field might cause fears of becoming subordinated to machines or even becoming obsolete. You certainly considered such fears when thinking about this long-term goal?
KITANO: I have two comments on that. We know that computers are better than humans in some aspects. But we donít feel subordinated. We accept that they have superior abilities in all these number-crunching tasks, even in chess. After all, a computer is just a machine that humans can utilize. The same thing can be said about robots. I donít think people will become feeling subordinated to them, even if robots will be able to beat the human soccer world champion. It may well be, though, that by 2050 we have mixed teams of humans and robots competing for the soccer world cup. That is what RoboCup really is about: Not the confrontation humans versus robots but the collaboration between them. Itís about how we can best utilize this technology.
QUESTION: That brings me to another point: Since neuroprothetics is an important application of robotics technology, humans and robots might have become hardly distinguishable by 2050. There may be human players with artificial legs or eyes and robots with organic components. Such developments might compromise the long-term goal of beating the human soccer world champion.
KITANO: Yes, in such case I would say the technology is so advanced that the aim of RoboCup is already met. If the initial goal became obsolete in this way, that would be just fine with me.
QUESTION: Can you sketch out the roadmap to the long-term goal? Which main milestones do you see on the way to 2050?
KITANO: In 2002 we introduced a remarkable change in the middle size league by removing the wall and saw that the teams were still playing just fine. We are discussing about going further ahead in this direction to a more conventional soccer environment for instance by using natural ambient light, enlarging the field, and using more players. Also, we are thinking about exchanging the orange ball against a regular soccer ball. These changes will probably take place during the next five years. What is more interesting is the question when we will have a humanoid robot that can run and jump. Right now they only can walk. I think running can be accomplished quite quickly, within a year or so. But the next task will be to make this movement pattern stable and robust. That will probably require a change of the basic mechanical structure from the gear and motor system to something like artificial muscles, which will have a major impact in a broad range of industries. A second aspect is how we can change the material we use in the construction of robots. Metals and plastics that are used now are very hard. We want to switch to softer materials. The third aspect is a more robust artificial intelligence. Currently, the intelligence of robots is by far not sufficient to play soccer against humans. Finally, the fourth aspect is energy supply. We need very advanced fuel cell batteries or probably something more futuristic to keep robots moving, running and jumping over 45 minutes time. Itís hard to predict when and how these problems might be solved but some major breakthroughs very likely will occur between 10 and 30 years from now. That will be the most exciting time. After that we will have to integrate all these different technologies.
QUESTION: What were your personal Ñgreat momentsì of RoboCup during the first years?
KITANO: Well, it was very moving in year one when we had announced RoboCup for the first time and saw that teams were actually arriving, bringing their robots with them. Before that I wasnít quite sure if all would work out as planned. But now I saw: RoboCup is going to be real! That was very exciting. Another exciting moment was in 2002 at the RoboCup world championship in Fukuoka when I saw 120.000 people coming into the baseball stadium to see the robots. Only five years ago we had started in a tiny corner of a conference hall to grow up to such dimensions in such a short time. To me as an organizer these were the two most exciting moments. Also, it was very moving when Robin Murphy deployed her rescue robots at the remains of the World Trade Center. It showed me that RoboCup really had an impact on society and that we were doing the right thing.
QUESTION: Do you expect RoboCup technology to influence human soccer one day, for instance by using simulation in the preparation of human games?
KITANO: That is very possible, although I canít predict when. You might get statistical data of your opponent from video recordings, do some analysis, and develop counter measures. In some sports like formula 1 there is extensive simulation going on already. There also have been some contacts between RoboCup and human soccer with the Japanese J-League officially supporting the RoboCup world championship 2002. But these thing are in an early stage and are developing slowly at the moment.
QUESTION: In Germany a new system is under development to record the movements of soccer players by using transmitters attached to their bodies. These data might be of use for improving training, transmitting games on the internet, or even by feeding them into software agents and looking if a change of players at certain moments could have led to a different outcome.
KITANO: That is very interesting. In RoboCup we already have teams, especially in the simulation league, that learn from their opponents behavior. We also have the coach competition with a simulated coach analyzing the game and advising the players in real time. But to apply that technology to human soccer we were lacking the data from human games. Such system might change that and lead to an application of RoboCup technology to human soccer quite quickly.
QUESTION: Did you observe different robot cultures in the many participating countries? Especially in Europe we are fascinated by the enthusiasm the Japanese people show for robots but itís hard to find an explanation for these differences.
KITANO: Basically, the Japanese love electric toys like the walkman or the play station. I donít know why, it seems to be rooted in our culture. Concerning robots especially, we have this culture of cartoons and animations like ìAstro Boyî and other robot stories which now exist for more than 40 years. Some researchers have actually been motivated to study robotics by looking at ìAstro Boyî. But now we have such influences outside Japan, too, with very popular movies like ìStar Warsî where the robots are the good guys. When I talk to participants at RoboCup and ask why they chose to study robotics they often tell me that they have seen ìStar Warsî when they were kids and have been fascinated by C3PO and R2D2.
QUESTION: To play soccer against humans robots certainly have to be humanoid in shape. But what else do we need humanoid robots for? How do you justify this huge effort to build biped robots?
KITANO: Of course, robots to be used in the office or in factories donít need to be humanoid. Even biped robots donít need to be humanoid in shape. But the humanoid appearance has a symbolic significance that is especially important in entertainment and has some relevance in other areas of human-robot interaction, too. Also, if we want to use robots for research in human behavior and motion they have to be humanoid in shape. Some neuroscientists already use robots to better understand human motion control. Considering more practical applications like neuroprothetics that you already mentioned, we also need to work with humanoid robots that can run and jump if we want to able to replace damaged parts of the human body with artificial limbs. Therefore I think that we should continue research in humanoid robots even if they could not be used for industrial or commercial purposes. Some of the technologies that we develop on humanoid robots will be of use on other types of robots, anyway.
QUESTION: You are also known as a founder of the new science discipline of systems biology. Does that have relations to RoboCup?
KITANO: I consider systems biology and RoboCup as completely separate scientific projects. Although, when you go into the very essence of robotics research and biology youíll find they have something in common. Both are dealing with very complex systems. Both are trying to understand how such systems can become robust against environmental frustrations and internal disruptions and still express certain functionalities in a stable manner. So there are some similarities, although robotics tend to focus more on practical applications while systems biology mainly deals with fundamental issues. Some time in the future they may work together more closely, though, for instance when we try to apply genetically modified tissues to mechanical systems. But that is probably still 20 years from now.
QUESTION: To be able to construct robots we have to learn from nature, but by building them we also learn new things about nature?
KITANO: I think so.
QUESTION: In human soccer we often attribute the players with having a good Ñgoal instinctì or a good feeling for the ball. How much feelings and emotions do soccer robots need and how do you implement emotions in artificial intelligence?
KITANO: That is an interesting question. I donít know if we really need emotions in a robot soccer player. It may not be necessary. One of the interesting episodes from the chess competition between Anatoly Karpov and Deep Blue was a moment when Karpov was shocked and psychologically damaged by a very unusual move of the machine. He lost the game because of this psychological problem. Deep Blue didnít have such problems. That made up part of its strength. Such considerations may apply to soccer, too. If you make a mistake you worry about it and that can affect the play. It could be an advantage for the robots not to worry about mistakes. But these questions are an interesting topic for further research.
QUESTION: In humans emotions are a very effective means of information processing. Robots might also need different levels of information processing, similar, though not equal to human emotions.
KITANO: Yes, but a faster processor might do that, too.
QUESTION: Will robots be granted political rights one day in the future?
KITANO: I donít think so. I doubt if we ever want to build that kind of objects. It may be technically feasible some time in the future, but probably not socially or politically. There will be a lot of objections against building such devices.
QUESTION: Could you imagine robots sitting in the audience in the year 2050 and getting excited while watching their team playing soccer?
KITANO: O yes, that I can imagine, absolutely.