Autonomous driving startup Turing is aiming to achieve fully autonomous driving and is using AI with higher performance than GPUs as in-vehicle AI. It uses an end-to-end AI model running on NVIDIA Orin and has successfully completed test runs in 5 cities across Japan.
Founded in 2021, autonomous driving startup Turing is aiming to achieve the world’s first fully autonomous driving (Level 5). The company places the highest priority on AI performance and has adopted a strategy of enabling urban driving through improvements in AI performance, rather than by following the conventional approach of increasing the number of cameras and sensors.
The vehicle is a modified commercially available Toyota Alphard minivan, supported by a total of 7 cameras, comprising 2 front-facing, 4 side-facing and 1 rear-facing camera, along with basic sensors and an in-vehicle computer. The AI uses an end-to-end architecture in which a single model processes everything from input to output.
Large volumes of driving data are used to train the AI. To overcome the limitations of simply increasing the amount of data and the number of GPUs, the company has adopted a design that enables real-time decision-making and control inside the vehicle. The driving model is limited to approximately 100 million parameters.
The company uses the in-vehicle NVIDIA Orin chip for large-scale data training and its computing infrastructure. It has already successfully completed autonomous driving test runs in 5 cities across Japan, and development is continuing toward future commercialization.