Turing Achieves Japan’s First Real-Time Autonomous Driving on Public Roads Using a VLA Model; Releases “RACER” Causal Reasoning Dataset and “DriveTiTok” Image Tokenizer
Turing Inc. (Shinagawa, Tokyo; CEO: Issei Yamamoto), which develops Physical AI for autonomous driving, today announced that it has become the first company in Japan (*1) to achieve real-time autonomous vehicle operation on public roads using a Vision-Language-Action (VLA) model.
In conjunction with this milestone, Turing has released “RACER,” a causal reasoning dataset for autonomous driving, as well as “DriveTiTok,” a specialized image tokenizer.
This development was carried out as part of the “GENIAC” (Generative AI Accelerator Challenge), a program supported by the Ministry of Economy, Trade and Industry (METI) and NEDO to strengthen Japan’s generative AI infrastructure. A portion of the dataset and pre-trained models is now available on Hugging Face. Turing is also sharing technical insights through its tech blog to contribute to the advancement of autonomous driving across both industry and academia.
*1: Based on Turing’s internal research as of March 2026, regarding domestic cases of autonomous driving involving real-time inference and vehicle control on public roads using a VLA model.
Real-Time Control via VLA Model
The VLA model integrates visual data from cameras with language-based contextual understanding to predict and execute driving actions such as steering, acceleration, and braking. Unlike traditional End-to-End (E2E) models that rely primarily on raw image and sensor inputs, Turing’s architecture incorporates a Large Language Model (LLM) backbone to enable integrated decision-making.
Turing independently developed and trained a VLA model with approximately 2 billion parameters and optimized it for on-vehicle computing environments. The system demonstrated stable autonomous driving performance on public roads, achieving real-time inference and control at 10 Hz.
Since 2023, Turing has focused on research and development of LLM-based autonomous driving. This achievement represents a major step toward realizing “Physical AI” developed in Japan, and Turing will continue to accelerate both technological innovation and real-world deployment.
Tech Blog：https://zenn.dev/turing_motors/articles/f5e44178d78153
< YouTube：VLA Model DriveHeron / チューリング株式会社 >
Causal Reasoning Dataset: “RACER”
“RACER” (Rationale-Aware Captioning of Edge-Case Driving Scenarios) is a dataset designed to enhance the causal reasoning capabilities required for advanced VLA models. By capturing the rationale behind driving decisions through structured causal descriptions, it enables AI systems to better understand why specific actions are taken, leading to more appropriate and reliable behavior.
Hugging Face：https://huggingface.co/datasets/turing-motors/RACER-Mini
Tech Blog：https://zenn.dev/turing_motors/articles/aeb2b1219eb6ee
Image Tokenizer: “DriveTiTok”
“DriveTiTok” is an image tokenizer that converts driving video into discrete tokens, compressing data to approximately one-hundredth of its original size. By incorporating temporal dynamics and scene context, and leveraging information from preceding frames, it enables efficient compression while preserving the critical visual information necessary for driving decisions.
Hugging Face：https://huggingface.co/turing-motors/drive_titok_s_640
The results presented in this press release are based on outcomes achieved with support from the “GENIAC” (Generative AI Accelerator Challenge), a project led by the Ministry of Economy, Trade and Industry (METI) and the New Energy and Industrial Technology Development Organization (NEDO) to strengthen Japan’s capabilities in generative AI development.
Reference Press Release：https://tur.ing/news/20250715/
Turing will continue to advance cutting-edge research and development in Physical AI for autonomous driving, with the goal of realizing full autonomous driving.
About Turing
Turing is a deep-tech startup dedicated to the development of fully autonomous driving. We are simultaneously developing End-to-End (E2E) autonomous driving AI—which integrates environmental perception, path planning, and vehicle control into a single system—alongside Large Foundation Models capable of understanding human social norms, backgrounds, and contexts. By integrating these core technologies, Turing aims to achieve “fully autonomous driving,” where vehicles can operate on behalf of humans under all conditions.
Company Overview
Company Name: Turing Inc.（Japanese：チューリング）
Headquarters: East Tower 4F, Gate City Ohsaki, 1-11-2 Ohsaki, Shinagawa-ku, Tokyo
CEO: Issei Yamamoto
Founded: August 2021
Business: Development of fully autonomous driving technologies
Website: https://tur.ing/
Careers
Turing is hiring individuals passionate about transforming the world through fully autonomous Japanese driving technologies. Please visit our careers page to learn more. We also host regular events such as open office sessions and Tech Talks.
Media Contact
PR Representative (Abe)：pr@turing-motors.com