Q&A with Hideki Kozima - How Keepon was born and what comes next

Since we reported about the commercialization of Keepon by BeatBots, we heard/read quite a number of comments regarding its cost and also some concern from the academic world that the dancing aspect of the robot is getting more attention than the "serious" research being done with it. 

We heard you! We've asked Hideki Kozima, the developer of Keepon, to answer these important questions. This Q&A article, originally in Japanese, has been edited and translated into English. 

 
(Photo:Hideki Kozima with Keepon and Infanoid)

Q. After people heard about the news of Keepon being commercialized, some readers asked why the robot had to be so expensive. Why does Keepon Pro cost $30,000?

A. The short answer is the high cost of manufacturing in small volumes and the high-end electronic components that we use.

Since Keepon was initially developed as a tool to conduct psychological experiments to observe voluntary communication behaviors between children and the robot, safety was of paramount importance. Children will not only touch Keepon but they will hit it, sit on it, bite it, etc. The robot had to be safe and also resilient enough that it wouldn’t break easily.

So Keepon’s body is made from silicon rubber, and while there is actually nothing complicated inside the body, the mechanism and the electronics are housed inside the black cylindrical base, and the body is driven by wires. This mechanism not only enables safety and compactness, but it also allows Keepon to make life-like movements. Although the four motors are moving separately within, from the outside it looks like the robot is moving as a whole, and I think it is this movement that has attracted so many people around the world.  But the flip-side of this is that the hand-machined mechanical parts are expensive to manufacture.  Furthermore, our manufacturer’s (Kokoro’s) experience with using silicon rubber for robots has contributed a great deal to making Keepon, but it is still time- and labor-intensive for them to handcraft the bodies one by one.  Most of Keepon Pro’s cost goes towards manufacturing: machining, 3D printing, molding, and so on. 

In addition to manufacturing costs, we use miniaturized and high-quality components.  Keepon has two CCD cameras made by ELMO that cost several thousand dollars each with optics, a high-end Sony microphone, and four actuators manufactured by Maxon of Switzerland that are several hundred dollars each.

I developed the controlling module - both hardware and software - from scratch so that it can fit inside Keepon’s small body. Keepon uses Pico-2 (motion controller) and Poco-14 (motor driver) boards, and since they can only be produced in small lots, that adds extra cost. By the way, we’ve decided to open up the specification of these controllers under the name ClayBot so that people can use them to build their own robots, and I can talk more about this later.

So, to summarize, Keepon was initially developed as a tool for academic research and thus it required expensive components to conduct accurate experiments.  When I first designed it in 2002, I never dreamed that it would be commercialized. It never occurred to me that people would want it until Marek posted a video on YouTube in spring of 2007.

Q. But you are planning on producing a cheaper version of Keepon.

A. Yes. Our first product, Keepon Pro (which has the same specifications as the original Keepon), costs $30,000. But we are committed to making a more inexpensive model.  The newer version will not need the precision to conduct psychological experiments, so the goal is to streamline the manufacturing process without losing the original life-like movements and presence. We are designing a new mechanism from scratch.  Customers such as schools for children with disabilities and elderly care facilities will be able to utilize it for their needs.

Q. How did you come to develop Keepon?

A. First let me go back a little bit to how I became interested in robotics.

Since I was in college, I have been interested in developing a computer that can learn language, so I majored in computational linguistics. Computational linguistics is a field in which you do research on how to make computers understand human natural language. I chose this field because I thought that the process of a child learning a language is the pillar of human intelligence, something only humans can achieve. In other words, by studying how humans understand and produce language, computational linguistics is the study of a “human science.” I feel that robotics is similar, in the sense that I am studying humans by using robots as a mirror.

Well, in the late 1980s, MIT’s Rodney Brooks and others introduced the “embodiment theory” into the field of artificial intelligence (AI), claiming that true AI can only be achieved by machines that are connected to the world through a body. I realized that I wouldn’t be able to fulfill my dream of making a computer that can truly understand words such as “tired,” “heavy,” or “cold” without the machine being able to experience these situations through a body. This realization led me into robotics, and I was fortunate enough to have the chance to do research at Prof. Brooks’s lab at MIT in 1998-99.   

My motivation in research has always been the desire to understand human communication, and the developmental process behind language, so that I can recreate that model in machines. It was in this flow of research that I encountered autism. Autism is a disability in verbal and nonverbal communication skills. By understanding autism, we can gain knowledge of the normal cognitive processes behind communication. So I started building robots as a platform to gain this knowledge and to demonstrate my findings.

In the year 2000, I developed Infanoid - a child-size upper-torso humanoid robot - to conduct research on physical communicative behaviors such as joint attention, eye-contact, pointing, and emotional expression. But I found that many small children under the age of 4 were scared of Infanoid. They were too distracted by the lights, sounds, and movements of Infanoid that they couldn’t grasp the holistic meaning of what the robot was trying to do.

So I decided to design a simpler robot in order for children to be able to intuitively understand the robot’s intentions and expressions.  I designed Keepon in mid-2002, and with the help of Kokoro engineers, Keepon was born in March of 2003.

Q. Please tell us about your research with autistic children and Keepon.

A. In general, people think that autistic children do not interact with others because they have no motivation to do so. But I have a different view. I think that autistic children actually do want to interact with the outside world, but can’t.

Autistic children will generally be scared of Keepon at first, but they are also very curious. After several months they will understand that Keepon will not harm them, and then we will see them start to voluntarily touch and make eye-contact with it. Although the process differs by individual, almost all autistic children come to enjoy their interactions with Keepon. Autistic children who have difficulty interacting with other people can become friends with Keepon.

Why? I think it’s because Keepon is simple.

Humans are complicated. Typically-developing children are able to interpret the emotion and attention of others from the complicated movements that humans make. But to autistic children, the complicated movements become a deluge of information that is difficult for them to process.  They can’t find meaning in the enormous amount of information.

Although Keepon is simple, it can make eye-contact with people and express emotion by bobbing up and down and cocking its head. I think that autistic children are able to process these simple movements, understand Keepon’s internal state, and interact with Keepon more easily.

This is a relatively new model of autism, in line with recent thinking on the sensory and perceptual characteristics of the disorder, and I would like to confirm my hypothesis through field research. And I am happy to say that we are gathering scientific data to back up my theory.

And my hope is that Keepon will be useful outside of research too. Human beings obtain social skills by interacting with other humans. If Keepon can become a trigger for autistic children to experience social interaction, then it might support their social development. If a robot can become a catalyst between the child’s natural growth and desire to interact, and the power of people supporting the child (such as parents and therapists), that would be great. That would be the ideal form of technology.

Q. Will your research be useful in developing robots of the future?

A. Yes, I think so. Right now, robots can’t understand human gestures and subtle nuances, so they are unable to communicate with people. The chasm between machines and humans is similar to that between autistic children and normal children. So I believe that if we can develop an information processing model for autistic children, we will be able to use the results to develop robots that can communicate better with people.

Q. Tell us about your ClayBot initiative.

A. This is a sub-project still under development, but the basic idea is to open up the specifications of the controller and motor drivers that I developed for Infanoid and Keepon under the name ClayBot, so that anyone can use the circuits and firmware in their own projects.

The reason I decided to open up ClayBot was something Marek told me. He said that there are two types of graduate students working on HRI (human-robot interaction)  research: one group uses commercially available robots to conduct research, and the other tries to develop their own original robots.  The latter group spends a great deal of their time in graduate school designing and trying to make the robots work before they can start working on the interaction research. I would be more than happy if they could use ClayBot to increase their productivity and to realize their robots and ideas more easily.

It is especially hard to use off-the-shelf components when you are developing a small robot.  Commercially available components are too large to fit into a limited amount of space. With ClayBot, the motion controller board and the motor drivers are all based on 60mmx40mm circuit boards. 

Q. You and Marek have founded a company called BeatBots. How do you feel about stepping into the business world?

A. I consider myself a social entrepreneur. I would like this venture to succeed not just as a commercial enterprise but also as a social one.

My hope is that we can expand the use of robots in what I call “communication care.”  That is, to utilize robotics and interface technologies to support people who have difficulties in communication for one reason or another. That would include people with autism, Alzheimer’s disease, aphasia, and amnesia. Also, from a broader perspective, communication care applies to fostering social intelligence in typically-developing children, maintaining long-distance relationships, and diminishing the “communication divide” between people who are good at social communication and those who are not as good at it.

By expanding Keepon’s reach and presence in the research, hobbyist, and art communities, we aim to make robots like Keepon more accessible to the care facilities and schools that have a real need for communication care. And I really hope that people will come up with other uses for Keepon.

We plan to come out with other products, but our basic stance is to develop products and design concepts that increase the well-being of people and relieve the stress of our everyday lives.