Shinpei Kato: Leading the Development of an OS for Autonomous Driving

Autonomous driving is a world that looks different depending on your field of expertise. Most people probably imagine the shape of a car, but from a computer science perspective, it is closer to a robot. To me, it is an image of numerous computers and sensors connected via a network. It doesn't feel like I'm building a car. Software development plays a vital role in autonomous driving technology.

At TIER IV, we have led the development of Autoware, but when we say "OS," what that refers to is surprisingly ambiguous. For example, "Windows" is a general term that includes software like Office. Autoware is similar to this, but an OS is ultimately just a platform and is not something used in isolation. The current situation is that various people are putting different things on top of it and testing them.

Various experts are involved in autonomous driving, and I'm sure people in each field feel the same way. Truly diverse perspectives are involved.

My area of expertise is computer science, but putting autonomous driving into practical use is, from a business perspective, on a scale equivalent to starting another company like Microsoft. It is a difficult world where specialists are involved in each individual technology, yet the business won't succeed unless you have all fields in mind.

To put autonomous driving into practice, various things must be added to the OS. Creating just the OS is possible if you have computer science knowledge, but the hard part is coordinating the interfaces with other fields when adding something. Therefore, to incorporate the voices of many experts, we launched an international industry organization called the Autoware Foundation in 2018.

Autoware is being developed using an open-source framework where anyone can obtain the source code, which serves as the program's blueprint. Therefore, many engineers other than TIER IV employees are involved in development. The Autoware Foundation is the organization that manages this open-source development.

In addition to TIER IV, more than 70 businesses and organizations participate in the Foundation, and they serve as the core members of development and decision-making. TIER IV is one of those members. I serve as the Chairman, but the Foundation includes people from various backgrounds, from corporations to organizations and research institutions like universities. The number of members joining Slack reaches about 3,000 to 4,000, and the Foundation manages all of them.

I think it's about 20 countries. Of these, about a thousand people are likely involved in the actual work of writing program code.

The development of Autoware began at Nagoya University in 2013. Immediately after starting, we received a huge response from all over, and we founded TIER IV in 2015. For the first two years or so, it was operated by TIER IV, but I realized that open source does not show its true value if it belongs to a single company, so I launched the Foundation. With an industry organization as the parent body, the system ensures that even if TIER IV were to withdraw, development would continue through other participating companies.

Directly, it is open to developers. But as a result, isn't it everyone, including users? I believe that by allowing anyone to develop, autonomous driving will reach almost everyone. By democratizing technology, many people can enjoy its benefits. I believe this is the state of highest value.

The "IV" in "TIER IV" also stands for "Intelligent Vehicle." In the automotive industry, which consists of a multi-layered subcontracting pyramid structure, there is a custom of categorizing companies into Tier 1, Tier 2, and Tier 3. However, when it comes to the development of autonomous driving, I believe all these hierarchies will disappear.

That said, automobile manufacturing will not become a disorganized horizontal division of labor. In product development that emphasizes safety and quality, such as automobiles, it is necessary to vertically integrate each supplier. A structure that opens up this vertical integration ecosystem and incorporates new technologies like autonomous driving is important.

That's right. In that sense, my basic thinking is that it would be unthinkable to implement a horizontal division of labor for the parts that are already saturated.

First, as a researcher, I felt curious about the technical challenges. Furthermore, autonomous driving has social and industrial value. For me, the fact that these three factors were aligned was significant.

I wanted to make this into a form that anyone could do, so I open-sourced it. Google was already advancing the development of an autonomous driving OS at the time, and I felt that simply following them would not allow us to overtake them quickly.

Yes. When I was in high school, I read a book by Bill Gates and decided to study programming. Following Gates' example, at the Yagami Campus, I joined the laboratory led by Professor Yūichirō Anzai (which later became independent as Professor Nobuyuki Yamasaki's laboratory) and built OSs. Including part-time jobs, I have built about 20 OSs so far.

After that, my interest shifted toward supercomputer research. After submitting my doctoral thesis at Keio, I moved to the University of Tokyo and continued my research as a researcher under Professor Ishikawa. Supercomputer research was interesting, but I gradually began to want to be involved in things that lead to solving social issues, so I shifted my focus to developing an OS for autonomous driving, which is an adjacent field to robotics.

I studied at Carnegie Mellon University in 2009. Before that, I was working with the National Institute of Advanced Industrial Science and Technology (AIST) and Professor Ishikawa on research into "real-time distributed information processing for humanoid robots." Fortunately, while studying abroad, I learned that the same technology was being used for autonomous driving and became interested. For example, I saw the potential to solve social issues such as traffic accidents and labor shortages in logistics. Furthermore, there was an existing giant market in the automotive industry. My motivation for development grew within me.

When I started developing Autoware at Nagoya University after returning to Japan, I already felt that Carnegie Mellon University's autonomous driving technology was insufficient from a computer science perspective, and at the same time, I saw potential in open-sourcing it.

That's right. By applying humanoid technology, we initially put software for humanoid robots developed at AIST directly into cars.

Yes. There is no trace of it now, but the origin is the HRP software.

My image at the time was software development, not an OS. I decided to rebuild the OS (in a broad sense) based on the prototype. As a result, the functions were enhanced. What I worked hardest on was defining interfaces and designing the architecture using a software platform for robot development called ROS (Robot Operating System).

In addition to students from Nagoya University, we had engineers from AIST and other places join us. It was a few people at first, but eventually, it was about 20 people in total. It was also significant that Nagoya University was selected for the Japan Society for the Promotion of Science (JSPS) Program for Leading Graduate Schools and the Japan Science and Technology Agency (JST) Center of Innovation (COI) program. All the students were employed by those programs and were in an environment where they could focus on their research. Thanks to that, we were able to work on the development of Autoware over several years.

The reaction was huge. We received inquiries from automakers like Toyota, electronics manufacturers like Sony, and many other companies.

Even after founding TIER IV, inquiries from overseas companies and others never ceased, so I decided to transfer ownership to a non-profit organization and launched the Autoware Foundation in 2018.

It's difficult to define what counts as "adoption," but it's about 1,000 companies. Among them, there are more than 10 partners who are actually integrating it into vehicles with an eye toward market launch. On the other hand, it is being utilized in various ways, such as being used for development benchmarks (indicators) or education. When we held a workshop called TIER IV Academy for about three years, we had applications from over 200 companies. I have the impression it is spreading quite widely.

For example, it has been adopted by several Japanese automakers. Autoware is used in many of the autonomous driving demonstration tests being conducted in Japan. Some companies have already begun using it commercially for transport within factories.

This is a very good situation for me. What TIER IV wants to do is not mass-produce vehicles, but to mass-produce companies that mass-produce vehicles using Autoware. TIER IV's concept is that anyone can become a manufacturer if they use Autoware. People in Japanese startups are also contributing a lot to feedback on open source, so we want to actively support them as well.

TIER IV's revenue model and open-sourcing are separate matters. For someone who obtains software from a customer's standpoint, it hardly matters whether it is open source or not. What is important for being chosen is being able to use high-quality products at low cost with peace of mind. Open source can only take development to about the halfway point. Beyond that is the work of TIER IV as a distributor.

TIER IV has several revenue models. One is the consulting business. we provide consulting for customers who do not know about TIER IV or autonomous driving. Second, there is a customer base that wants to do autonomous driving but doesn't know what to build. For those people, we provide something they can sell, saying, "Why not use this to find your requirements?" This is the so-called turnkey business.

The third is the vehicle development business, which provides vehicle development know-how to people who have the will to "create this kind of vehicle." The fourth is the partnership business (DevOps business) with so-called "Small TIER IVs" who handle platform building on our behalf.

Fifth, there is the AD (Autonomous Driving) software business, which sells software to customers with vehicle development capabilities. TIER IV provides various services like these.

It would have to be the ability to shorten the development period. There weren't that many cases where automakers were advancing in-house development of autonomous driving systems, and even if they try to advance in-house development now, it will incur a new and large cost burden, so a cautious decision is necessary. The people who handle this are those in startups. And many startup members are enrolled in the Autoware Foundation. TIER IV's mission is to provide a platform that anyone can use for people who want to develop autonomous driving. Therefore, as long as the installation of Autoware remains the standard, I believe TIER IV and the Autoware Foundation will continue to be at the upstream of development.

TIER IV is currently operating with a scale of 300 to 400 people, but other companies at the core do not need to be of an equivalent scale. This is because TIER IV can provide the platform.

About once every two months. I think it's a high pace, but we are able to release quickly because we have adopted agile development methods that repeat in small cycles of functional units in the development process.

We are looking ahead to around 2030. When autonomous driving technology is put into practical use, it may become possible to use various vehicles like taxis. When considering the practical application of "robot taxis" with no one in the driver's seat, we are drawing a roadmap that far ahead.

It is expected to be around 2030. It is already possible to run them now, but they are not yet profitable, and legal aspects must also be cleared. Those things will be in place around 2030. As a concrete image, we are thinking of practical use in rural areas or overseas where there is no public transportation.

TIER IV's mission includes "Creation and Destruction," and while the development of Autoware is of course creation, we must also be sensitive to destruction. For example, if robot taxis become necessary in Tokyo due to labor shortages, the existing taxi industry will be forced to reorganize. But looking at it differently, you could say that the taxi industry is not functioning at that point. At that time, we will take responsibility for the robot taxi business. I believe such things are also part of TIER IV's future mission.