My History
Our company is a startup developing autonomous driving AI. We are taking on the challenge of achieving fully autonomous driving under all conditions. There are generally two main approaches to research and development in autonomous driving. The first is the modular approach. This method divides the autonomous driving system into several stages—or modules—such as “object recognition,” “motion prediction,” and “behavior planning.” Waymo, an autonomous driving company under Google, has based its development on this approach. Its advantage is that it allows easier analysis of where errors occur when problems arise. However, the interconnections between modules are complex, making it difficult to optimize the entire system simultaneously.
The other approach is End-to-End (E2E). In this method, sensor data is directly input into an AI model, which outputs steering, acceleration, and braking commands all at once. By training the AI on a massive amount of driving data, it is believed that it can make decisions similar to those of a human driver. It is an innovative approach that essentially means entrusting all driving decisions to AI. While the modular approach had long been the mainstream, Tesla, the leading U.S. EV manufacturer, adopted E2E, prompting an increasing number of autonomous driving companies worldwide to follow suit.
Since our founding in 2021, we have consistently adopted the E2E approach. Our ability to lead in this field ahead of others stems from my own background. During my time at the University of Tokyo, I began studying programming after repeating a year. As a member of the university’s Shogi (Japanese chess) club, I decided to develop a Shogi program, which I named Ponanza. After training Ponanza on data from 800 billion AI-vs-AI matches, it defeated the reigning Meijin, Sato, achieving the historic milestone of becoming the first AI ever to beat a human Meijin.
This approach—entrusting the learning of Shogi to AI—was in essence an E2E model. Although it was within the limited domain of Shogi, E2E enabled AI to surpass human ability. Inspired by this achievement, I founded Turing with the vision of applying the same concept to autonomous driving. At the time, the E2E concept was not yet widely recognized, and some investors were skeptical. However, given the direction of the industry today, I am confident that our decision was the right one.
The Present
Japan is one of the world’s leading automotive nations. It is home to major manufacturers such as Toyota and is globally recognized for its excellence in hardware fields such as vehicle design. However, in the area of autonomous driving—particularly on the software side—Japan lags behind other countries.
Several factors contribute to Japan’s weakness in the software domain of autonomous driving: 1) A shortage of professionals with advanced AI skills, 2) The difficulty of restructuring organizations to optimize software development, and 3) A lack of executives with sufficient AI literacy to properly allocate vast computing resources.
From this perspective, our company is uniquely equipped to overcome these challenges. My achievement with Ponanza—defeating the Meijin—brought me public attention and helped attract top engineering talent. Moreover, our team includes four Grandmasters from Kaggle, the world’s most prestigious machine learning competition, which naturally draws ambitious and highly skilled individuals who seek to grow alongside peers of the highest caliber.
Developing autonomous driving AI also requires enormous GPU and data center resources. In the U.S. and China, company leaders take significant risks and invest trillions of yen to secure such computing resources. Japan, by contrast, tends to underinvest in this area. However, because both I and other executives at Turing possess a high level of AI literacy, we have taken an aggressive stance toward investment in computational resources. Currently, with the support of government grants and research funding, we operate computing resources worth approximately 4 billion yen. We have also built and operate our own dedicated GPU cluster, significantly improving both development speed and cost efficiency. This top-tier development environment in Japan also helps us attract outstanding AI talent.
Under this framework, our current focus is the “Tokyo30” project—a goal to achieve over 30 minutes of fully autonomous driving on public roads in Tokyo by 2025. We have prepared more than 20 data collection vehicles equipped with cameras and onboard computers, and test drives are already underway.
The key to success lies in the quality of data collection. High-quality data means data gathered from driving that strictly adheres to traffic laws. Training AI on exemplary driving data leads to safer and more stable decision-making. To ensure this, we have invited a former police officer and ex-motorcycle patrol instructor to join the project. With such expertise, we are improving the accuracy of AI learning and building a foundation capable of responding flexibly to complex traffic conditions.
For the Future
The primary significance of pursuing fully autonomous driving lies in reducing traffic-related fatalities. This alone carries immense value for humanity and is one of the core motivations behind our challenge. However, our goal extends beyond that. We also see it as equally important to help Japan’s automotive industry maintain its global competitiveness in the years to come.
If our world-class autonomous driving AI can be combined with the safe and high-quality vehicle engineering that Japanese automakers are known for, we believe it could lead to a renewed global appreciation of Japanese automobiles. However, being a Japanese company does not automatically mean we can collaborate with domestic automakers. To become a trusted partner in building the future, we must first match and be recognized for our technological capabilities—catching up with leading international players such as Tesla and Waymo.
From a national security perspective as well, there is great significance in developing fully autonomous driving technologies domestically. If Japanese vehicles were to rely on foreign software, everyday driving data could flow abroad, and in times of crisis, vehicle control systems could become vulnerable. The spread of Japan-made autonomous driving software would therefore contribute not only technological value but also to the safety and security of society as a whole.
Just as the advent of the smartphone dramatically transformed society, the realization of fully autonomous driving will bring about a series of unforeseen changes. We believe that autonomous driving AI technology shares common ground with humanoid AI. Specifically, for AI to make behavioral decisions based on camera input, it must accurately understand its own size and shape. For a car, this means comprehending its width, length, and how much it moves in response to steering, acceleration, or braking. It must also understand the nature of objects ahead—bumping into a cardboard box may not matter, but colliding with a steel object could cause a serious accident. Modern AI systems are beginning to understand their physical dimensions and the effects of their actions, enabling them to grasp or manipulate objects based on this awareness. This mechanism is similar to how humanoids recognize and control their own bodies. Our mission goes beyond achieving fully autonomous driving. We will continue to deliver technologies—such as humanoid AI—that have a profound impact on society and help shape the future.