Interview Transcript
Q1: So we’d like to begin by asking a few questions about your background. If you could share your name, anything about your family or upbringing that you would like to share, schools you went to, and jobs you’ve held.
Kazuhito Yokoi: Okay. So my name is Kazuhito Yokoi. I’m a robotics researcher. In 1983, I entered the Tokyo Institute of Technology, and my advisor was a professor there at the time. I started my robotics career as an assistant. At that time, it was very different from today. It was the first time I was really tackling robotics. After I finished my master’s thesis, I moved into mechanical engineering robotics in Japan. I also worked on impedance control, or compliance control, of robotic arms. So I moved from working on legs to working on arms. After that, I had a chance to visit Stanford University, where Professor Khatib was my advisor. I worked with him during the period of mobile manipulation research. So at that point, I combined mobility and manipulation. After that, I returned to Japan, and Honda announced a humanoid robot. The Japanese government started a humanoid robotics project, and I joined that project at AIST. In 2000, mechanical engineering robotics and electrical technical robotics in Japan were combined into a new organization, AIST. At AIST, I worked with humanoid robots. We developed a series of humanoid robots, starting with HRP-1S, then HRP-2, HRP-3, HRP-4, and HRP-5.
Q2: So the first HRP-1 was developed by Honda?
Kazuhito Yokoi: Actually, at that time Honda was the leading company in the world. In our area, developing software within a humanoid robot, we called it HRP-1S. It was a kind of submodel of a humanoid robot. We developed software to control the humanoid robot body. We collaborated with other industries, and then we developed HRP-2 and HRP-3. In the project, we used HRP-1S. With HRP-2, we developed several tasks, including operating industrial vehicles like a backhoe with HRP-1S, and HRP-2 carrying panels together with humans. After the project, we continued to use HRP-2 and perform several tasks. For example, we studied motion in several kinds of environments. If there was a bar at a certain height, the humanoid could automatically crawl under the bar or go upstairs while using a side bar. We also developed the operating system of HRP-2. By automatically detecting objects, it was a perfectly model-based approach. If the humanoid knows a model of several items, it can operate in a program-based way. That was the focus of HRP-2 research. We also developed several motion-planning systems, but they were still model-based. After that, AIST developed HRP-3. We tried to use a humanoid robot in outdoor environments. That meant it was a dustproof and waterproof humanoid robot. Actually, it could not see, but it could work in the rain. After that, we developed HRP-4C. That was a large female-type humanoid robot. It could wear a wedding dress. So that was more in the entertainment area of humanoid robotics. We successfully continued humanoid robotics research. However, in 2011, we had a major event at the Fukushima nuclear plant. We discussed this with Honda. Honda had also developed ASIMO, but no humanoid robot could work inside a nuclear disaster area. So we discussed with Honda and developed a non-humanoid robot that could be placed inside the nuclear power plant and check conditions, especially in high places. Of course, there were several mobile-type robots, but legged robots and other robot types could not access those high areas. There were also some drones, but there was concern that nuclear dust containing radioactive materials could damage them. So that was another area we developed. After that, the DARPA Robotics Challenge was held, and I continued trying to participate in those kinds of projects. We developed HRP, meaning a more powerful robot. It could carry 20-kilogram panels and automatically handle those panels. This type of system could be used in home construction. I don’t know the exact English word, but something like a carpenter robot. So that is the history of humanoid development at AIST. Of course, after that we continued working on humanoid robots with Kawasaki Heavy Industries and other companies. Now I have moved out, so I cannot tell the details about later developments.
Q3: That leads me to a clarifying question. What is your current position or title, and where are you currently working, if you can share that?
Kazuhito Yokoi: I’m currently working with the National Institute of Energy and Industrial something. Sorry, I cannot remember the exact name.
Q4: It sounds like the trajectory of your career has largely been in industry. Is that true, or have you been mostly elsewhere?
Kazuhito Yokoi: No, no, no. Almost all of my career has been in a national research laboratory, AIST, the National Research Laboratory. Of course, we collaborated with some companies to make new robotics services and things like that.
Q5: Thank you. I want to jump back in time a little bit and ask about your earliest relationships with robots. First, how did you get interested in robotics in the first place? Were there books, films, or anything else that you came across?
Kazuhito Yokoi: Okay, okay. You mean when I was younger. I’m now over 60, so at this age, as a Japanese person and as a boy, the first time I became interested in robots was through Astro Boy.
Q6: So Astro Boy would be the first robot you ever heard of or encountered? And beyond just enjoying it, was that actually inspiring? Was it something that helped lead you to get into robotics professionally?
Kazuhito Yokoi: Yes, actually, after that, maybe around 1980, we had the Osaka Expo. There were some pavilions demonstrating robots. That was the second thing. It was a real robot, not animation.They were like manipulators from the 1980s. The 1980s were an important, epoch-making period for robotics in Japan. Of course, originally in the United States, there were advanced industrial robots, and Japan imported that technology. That became the foundation for two generations of industrial robots. I cannot remember exactly when PUMA appeared. PUMA is a robotic arm.
Q7: So you were mostly seeing this type of robot, like a robotic arm?
Kazuhito Yokoi: Okay. That is the reason why the Osaka Exposition was also important. Actually, the 1970 exposition was earlier. It was more like toy-type robots and things like that. I made my first robot at that time. It was not an actual robot in the modern sense, but in some way, as a junior high school student, I built something. After that, when I entered university, I really wanted to study robotics. However, my university did not have robotics, so I moved to the Tokyo Institute of Technology, one of the universities with robotics research.
Q8: You’ve already provided a nice picture of the arc of different robotics projects you’ve worked on. Could you share more detail about the first robotics project you worked on?
Kazuhito Yokoi: One of the early projects was a robot called TITAN. TIT stands for Tokyo Institute of Technology, and it was a Japanese working robot. I researched TITAN-3. Actually, in 1985, we had the Tsukuba Expo. In the Japanese Pavilion, we had three robots. One was TITAN, a crawler-type working robot. Another was a piano-playing robot, and there was another robot as well. That was a big challenge in Japanese robot history.
Q9: We’ve talked a little bit about the robot projects you’ve worked on throughout your career. I’d be interested in learning more about the research questions that have inspired you or guided your work. What have been some of the main research questions you’ve been interested in, and how have they changed over time?
Kazuhito Yokoi: So, again, what are some of the main research questions I have been interested in over the course of my career? I really wanted to make machines that could work automatically—something that could help humans. That was my main motivation. In the 1980s, there were several industrial robots, but they were only used inside factories. So I wanted to move robots outside of factories and into human environments. That was one of the reasons. Then I researched compliance control of robotic arms. This means softer control, or more compliant motions, which are helpful when a robot interacts with humans. For example, these are tasks where I do not want the robot to break things. That was the first direction. Also, those robots were fixed arms. When I went to Stanford, that was a great opportunity for me. In mobile manipulation research, the robot could move and also manipulate objects. It was closer to the type of robot I was interested in creating.
Q10: Would you say this interest in moving robots out of industrial contexts into more social or domestic contexts is related to your interest in humanoid robots over your career? Is there a concrete connection between that desire and the form of robots you’ve been interested in?
Kazuhito Yokoi: Yes, actually. When Honda presented the humanoid robot, almost everyone did not believe that a humanoid robot would become realistic like it is today. I also thought that a mobile manipulator was good enough to perform several tasks. We also worked with two mobile robots, Romeo and Juliet, which were two mobile manipulation robots. But then Honda announced the humanoid robot, and everyone changed their mind. It seemed like it might actually be possible. That is the reason why I also started to develop humanoid robot research.
Q11: What have been some major challenges you faced, whether technical, institutional, or social, that have arisen over the course of your career while pursuing those goals?
Kazuhito Yokoi: Humanoids have a lot of potential, but at the moment even just walking is difficult, especially outside of flat environments. Maybe walking on rough terrain or outdoor environments was something everyone in humanoid robotics expected robots to eventually do. Expectations became larger and larger, and maybe they are still the same today. But in reality, it is different. Even just walking is difficult at the moment. However, walking alone is not very useful. So we tried to perform several tasks, including manipulation. That is even more difficult. Actually, working in a factory is easier. Industrial robot arms are fixed, and the work environment is fixed. Everything is very deterministic and modeled, so precise position control is possible. But a humanoid robot is not fixed to the ground. The robot’s hand is also not fixed. So we need to observe the environment, measure position and orientation, and understand how to handle objects. After that, we need to generate motions automatically. That is very difficult.
Q12: Have there been any tools that have appeared during your career that have had a significant impact on your work or opened up new possibilities for achieving those goals? By tools, I mean anything—from Kinect, large language models, or specific hardware.
Kazuhito Yokoi: Yes. For humanoid research, and not only humanoid research but robotics research in general, hardware is one of the key elements. We can design any type of robot on paper. But making it real is more difficult. How do we achieve that? We are very fortunate to collaborate with companies. For example, Honda developed humanoid robots, and other industries also built very good hardware. We can then concentrate on the software side. That is a very important thing.
Q13: Can you talk a little bit about funding? Who has funded your research, and how did you develop those relationships?
Kazuhito Yokoi: Past humanoid projects were funded by the Japanese government, especially the Ministry of Economy, Trade and Industry. My current institute, AIST, also supports these projects. The national government supports humanoid robot projects. After that, of course, we continued to seek government funding, but we also collaborated with companies.
Q14: How do you develop those relationships with industry?
Kazuhito Yokoi: Actually, some industries want to introduce robot technology into their businesses. In those cases, we collaborate directly with companies, or we create teams with industrial companies to work on new developments.
Q15: Would you say that different funding sources, funders, and relationships with them have influenced the direction of your research? Did they change the questions you pursued or things you focused on?
Kazuhito Yokoi: Actually, we collaborated with CNRS. We created a joint robotics laboratory with two branches, one in Japan and one in France. I was the first director on the Japan side. That international collaboration was a very important thing. Especially our collaboration with France was significant. Several European research laboratories had the opportunity to work with humanoid robot technology. For example, LAAS and other institutes of technology became involved in humanoid robotics research. Through this, I developed relationships with many members of the research community. We were able to build a large community in humanoid robotics. We also started the Humanoid Robot Conference. That was a very important development.
Q16: I want to get to some specific questions about humanoids in a moment, but before we get there, I have a couple of questions. First, have there been things you have been interested in pursuing but haven’t had the chance to pursue over your career? Questions, projects, or anything like that?
Kazuhito Yokoi: I didn’t quite understand the meaning of your question.
Q17: The question is really: over your career, you have focused on certain areas. Have there been other things you were interested in pursuing but did not have the chance to, perhaps because you did not have funding, time, or resources?
Kazuhito Yokoi: I was lucky. I was able to do many of the things I wanted to do. Of course, AIST is a national laboratory, not a university. Later in my career, I moved into the management side. After that, I could not do robotics research anymore. That was not a good thing, actually. But before that, I was able to do what I wanted to do.
Q18: What would you say has been your most important contribution to robotics so far?
Kazuhito Yokoi: My—yes. Yes, I think it is humanoid robotics research. That is our big contribution to the robotics research community.
Q19: Is there anything specific you would like to expand on about the contributions you made to humanoid robotics or the humanoid robotics community that you think is worth noting?
Kazuhito Yokoi: Yes. Actually, we worked with Kawada Industry and other companies. Kawada Industry started renting humanoid robots to several Japanese universities and also to research centers in France. Because of that, several researchers were able to work with humanoid robots. Then researchers in Korea, the United States, and Europe tried to build their own humanoid robots. The first impact was in Japan, especially with Honda. But we expanded the impact to the world. That was a big contribution—not only from me, but from our entire team—to the robotics community.
Q20: On this topic, I’d like to dig a little deeper into the humanoid-specific material. When did you first begin working on humanoid robots specifically?
Kazuhito Yokoi: About 2000.
Q21: What kind of problem were you trying to solve at the time?
Kazuhito Yokoi: At that time, we were among the first to use an outside humanoid robot for many different tasks. Object recognition, environment recognition, motion planning, body motion planning, and operations with humanoids—everything was new.
Q22: Were there any specific publications that were significant that came out of that research?
Kazuhito Yokoi: Yes. Professor Hirai was a member of the humanoid research team at that time. His paper received a major prize. I forget the name. I think it became one of the most cited papers over ten years, maybe ten or twenty years. I also received several awards. That may have been partly because I took many videos of the humanoid robot.
Q23: Over the course of your career and across different decades, how would you say robotics has changed along the way? What were some important developments, and where do you see things going in the future?
Kazuhito Yokoi: I think the biggest thing is AI, especially generative AI and deep learning. That has been a very important change in robotics intelligence. Also, Chinese companies have developed very good actuators. That is another major change in robot hardware. So those two things are very important. Of course, there are also other developments, such as large GPUs and technologies that allow us to collect and process large amounts of internet data. But the two biggest changes are the intelligence side and the hardware side. The relationship between software and hardware is very important.
Q24: On that topic, can you expand a little bit on how you understand the relationship between a robot’s body and its controller?
Kazuhito Yokoi: The body is easy to understand, but the controller, in my sense, has several different levels of control. There is motor control, more advanced control levels, and higher-level control for performing tasks. Those levels are very different. At a high level, models and vision-language models are very helpful for allowing robots to enter different worlds and handle different levels of tasks. At the motion-control level, force control and dynamic movement are also important. However, we still need more details. Everyone is trying to obtain enough high-quality data for training and for vision-language models. But robots still need to physically contact objects. That means tactile information and force information are also important. Maybe sound information could also be helpful. We need more high-quality data. Hardware is also important, especially batteries. Battery life is still short. One company announced a self-charging battery system for a humanoid robot, but battery operation is still an important challenge.
Q25: Have those different challenges affected the way you have approached designing humanoid robots?
Kazuhito Yokoi: Yes. I worked with humanoid robots more than ten years ago, around 15 years ago. The difference between that time and now is very large, especially because of artificial intelligence. That is a major opportunity for the robotics community to achieve new developments in robot intelligence. Humanoid robots have one very good body. Not only humanoids, but all kinds of robots may reach a different level of intelligence.
Q26: You have talked about your personal history and the broader history of robotics that you have observed. Have you had any unexpected successes, failures, or mishaps over the course of your career, either from your own work or from other groups, that you would like to share?
Kazuhito Yokoi: Yes, actually, we usually present only success stories. But in reality, we had many failures—maybe ten times more failures than successes. I don’t know. Maybe everyone tries to hide failures because we do not publish them. So actually, we experienced more unsuccessful things than successful things. But from those unsuccessful things, we get one good success. That is the important point. Also, these days, young researchers need to understand the history of robotics. The same kinds of things happen again, although in different forms. It is very important to understand past mistakes.
Q27: Do you happen to remember any funny anecdotes or stories related to those kinds of mistakes or successes?
Kazuhito Yokoi: Actually, we did many demonstrations with humanoid robots. Sometimes, before a demonstration, we had major problems. We would fix the problem overnight, and then I would do the demonstration. Fortunately, we succeeded in the demonstrations. But every time we do something, new things happen. For example, in the DARPA Robotics Challenge, on the first day our robot fell down. It was not something completely unexpected—it was a known error in the system. That is what happens in the real world. A robot is just a machine, and software always includes bugs.
Q28: Moving back to a broader perspective, I have a few final questions about humanoid robots. In your opinion, what makes a robot a humanoid? There are robots in general, and there are humanoid robots. What differentiates humanoid robots from other kinds of robots? What makes them specifically humanoid?
Kazuhito Yokoi: That is actually a good question.
Q29: Do they need to have arms and legs? Do they have to have a face?
Kazuhito Yokoi: No, no, no. When we started the Humanoid Robot Conference, we discussed this deeply with our colleagues around the world: what is a humanoid robot? For example, is a biped walking robot a humanoid? I would say no. A Hollywood character that looks like a human is also human-like. So having legs or arms is not enough. Humanoids do not necessarily need to have those specific features. Of course, it should be something like a human, but it is difficult to make a precise definition. If an ordinary person looks at something and says, “That is a humanoid,” then we consider it a humanoid. So we include human-type robots in our definition.
Q30: So would you say a humanoid robot could have wheels, legs, or something else? Could it have one arm, two arms, or even six arms? What about something like a head or face? Are those necessary?
Kazuhito Yokoi: A face is not necessarily needed. I think we do not need that. Of course, we need something to look around the world, but we do not need a face. Facial expressions are not necessary. Of course, HRP-4C can create facial expressions and mouth movements, but I think that is not essential. The same applies to manipulation. Of course, if we want to manipulate objects, we may need arms. But humans can also manipulate objects in different ways. For example, a professional football player can use their foot to manipulate things.
Q31: I want to combine two questions here. You have mentioned that the fundamental interest in humanoid robotics came from shifting away from industrial robots designed for very specific circumstances toward robots that can operate in more diverse and open environments. Why pursue humanoid robots specifically for research? What is the significance of humanoid robots in this more open context? How do you see them being useful?
Kazuhito Yokoi: Actually, I work not only with humanoid robots but also with human-type problems. Humanoids are very helpful for understanding humans. That is my sense: understanding how humans think about and understand the world, and how humans manipulate the world. Why do we perform certain tasks? Humanoid robots help us study these questions. However, on the other hand, we also have mobile robots and drones. For some specialized tasks, a human shape is not the best solution. For example, a swimming ship is much better than a swimming humanoid robot.
Q32: Have you seen the results of your research applied in real-world situations? What does that look like? Has anything surprised you? The spirit of the question is about whether your research, which has mostly been in a research context, has moved into the broader world and been applied.
Kazuhito Yokoi: Research, in my opinion, is just research. We did not directly develop things for business. But we wanted to make good robots. Our hope is that robots help humans, create a better world, and contribute to a better society.
Q33: What would you recommend to students interested in robotics?
Kazuhito Yokoi: I always say that in humanoid development, especially humanoid robots, we should learn many different things. Not only mathematics or physics, but also painting, culture, philosophy, and law. Everything is actually connected. Of course, one person does not need to learn everything. But we should create teams that include not only engineering people, but also people from social fields. That is very important because robots work with humans. Robots enter human societies.
Q34: And the last question we like to ask is: was there anything we did not talk about that you would have liked to share, or any questions we should have asked that we did not?
Kazuhito Yokoi: Nothing special.
Interviewer: Okay. I want to make sure you have time to go get lunch, so I think we can end things right here if there is nothing else. Thank you very much.
Kazuhito Yokoi: Thank you.