By ATN Editorial Team
Summary: TIER IV has developed a new end-to-end architecture for Level 4+ autonomous driving, capable of handling previously unseen scenarios without human input. Starting in 2026, the architecture will be trialled across 50 locations in Japan to validate its real-world performance. Built on Autoware, TIER IV’s open-source platform, the system uses a hybrid of diffusion model-based AI and rule-based logic to deliver human-like driving behavior with robust safety and interpretability. The company is also leveraging simulation and synthetic data to accelerate training and improve model scalability for widespread deployment in mobility and logistics applications.
Auto Tech View
- End-to-end autonomy architecture uses diffusion models to imitate complex human driving behaviors like obstacle avoidance and busy intersection navigation.
- Hybrid AI and rule-based logic ensures both adaptability and deterministic safety, balancing innovation with operational reliability.
- Large-scale synthetic training data generation via simulation boosts scalability, enabling deployment across diverse commercial use cases.
TIER IV, the pioneering force behind the world’s first open-source software for autonomous driving, has developed an end-to-end architecture for Level 4+ autonomy, where no human intervention is present even in previously unencountered scenarios. Starting in early 2026, the architecture will be introduced gradually in mobility services at 50 locations across Japan as part of a large-scale demonstration to evaluate its real-world performance and capabilities.
Publicly available via Autoware*, open-source software for autonomous driving championed by TIER IV, the newly developed architecture applies diffusion model-based machine learning to a sequence of driving tasks, including prediction of surrounding objects and generation of vehicle trajectories. This allows the system to imitate human-like driving behavior, even in highly complex scenarios such as obstacle avoidance or turning at busy intersections. In parallel, rule-based components are integrated to ensure high interpretability and operational stability. This hybrid approach combines the adaptability of learning-based methods with the reliability of deterministic logic, positioning the architecture as a practical and promising foundation for Level 4+ autonomy.
To accelerate model development, TIER IV is taking advantage of Autoware’s modular architecture and simulation environment to automatically generate large-scale synthetic training data. Combined with real-world data, this approach has resulted in the efficient construction of high-performance models with both scalability and reliability.
TIER IV will continue to expand its training datasets and enhance model performance to further improve the robustness of the architecture. A variety of data-centric AI models will also be explored and integrated to ensure adaptability across a wide range of use cases, from privately owned cars to commercial vehicles for mobility and logistics services. Through the development of Level 4+ autonomy, TIER IV aims to help tackle major challenges in Japan, including the revitalization of regional communities and strengthening of industrial competitiveness.
For more autonomous vehicle technology news, click here.
