TIER IV, the pioneering force behind open-source software for autonomous driving, has developed software stacks for Level 4 autonomous driving powered by data-centric AI. Publicly available via Autoware*, open-source software for autonomous driving, the newly developed software stacks aim to expand operational design domains. Designed to be hardware-agnostic, they support various system-on-chip (SoC) and sensor configurations based on automotive industry requirements. By using TIER IV’s machine learning operations (MLOps) platform together with the new software stacks, automakers can continuously iterate and improve AI model performance. To validate the effectiveness of these capabilities, TIER IV has commenced testing through collaborations with partners including universities in three global hubs: Tokyo, Pittsburgh and Munich.
Software stack release
TIER IV has launched a Level 4+ initiative, envisioning the gradual expansion of fully autonomous driving into more complex environments. This approach starts with Level 4 under specific conditions and uses real-world operational data to continuously refine AI models and expand use cases.
The newly published data-centric AI technology is the core component supporting this concept, expanding the functional deployment of Autoware based on the end-to-end (E2E) architecture released in July 2025. The software configuration can be selected from two systems to ensure adaptability and scalability across diverse driving environments without hardware lock-in. This serves as the foundation for automakers to lead and internalize the development of autonomous driving systems customized for unique vehicle designs and use cases.
- Hybrid system with perception AI and planning AI: Uses diffusion models to probabilistically capture temporal changes in the surroundings. By combining this with environment perception from other ML models, it generates decision-making and trajectories, mimicking human driving behavior.
- E2E system: Treats the surroundings and driving status as vector representations. Leveraging the concept of world models, it integrates perception, planning and control into a single learning process, providing a seamless pipeline from environmental recognition to vehicle operation.
The software stacks are available on GitHub within the Autoware repositories. In collaboration with the Autoware Foundation, TIER IV aims to establish AI-based Level 4 autonomous driving as an industry standard by fostering a framework where academia, industry and the developer community can collectively improve the open-source software.
- Demonstration of Level 4 autonomous driving features in Tokyo
- Demonstration of Level 4 autonomous driving features in Pittsburgh
- Demonstration of Level 4 autonomous driving features in Munich
MLOps platform use
The MLOps platform delivered by TIER IV handles data-quality validation, anonymization, tagging for searchability and annotation based on assessments from active learning frameworks. It can also generate a diverse dataset by combining real-world and synthetic data to evaluate autonomous driving system functions in complex environments. These advanced technologies are sustained through collaborations with a wide range of partners, including the collaboration with the Matsuo Institute.
Looking ahead, TIER IV aims to realize highly practical AI-based Level 4 autonomous driving through collaborations with automakers, continuously improving AI model performance using large-scale driving data and various MLOps capabilities.
Testing in Japan, U.S. and Europe
To validate the effectiveness of data-centric AI for Level 4 autonomous driving, TIER IV is launching driving tests of Level 4 autonomous driving features in regions with distinct traffic characteristics, utilizing different vehicles, SoCs and sensor suites. Each test run lasts approximately 60 minutes. While a safety driver will be on board in accordance with local regulations, no manual intervention is expected under normal operating conditions.
- Tokyo: In collaboration with the University of Tokyo, Toyota JPN TAXI is used to evaluate the user experience when traveling between hubs in urban centers.
- Pittsburgh: In collaboration with Carnegie Mellon University (CMU), Hyundai IONIQ 5 is used for robotaxi tests in urban areas, including routes between Pittsburgh International Airport and CMU.
- Munich: In collaboration with the Technical University of Munich, Volkswagen T7 Multivan is used for safety evaluations across various urban driving scenarios in and around Munich.
Through an international framework built on the open-source ecosystem, TIER IV is committed to driving the deployment and sustainable evolution of Level 4 autonomous driving.
“To achieve Level 4+ autonomy, we need technology that evolves autonomously alongside the environments it serves,” said Shinpei Kato, founder and CEO of TIER IV. “Our new data-centric AI models and collaborative MLOps platform provide a common language and a shared foundation for the entire industry. By working with research institutions, industry leaders and the development community to advance autonomous driving technology through Autoware, we are creating an open, transparent environment that fosters continuous, collective innovation for the benefit of society.”
“Autoware serves as the global foundation where researchers, corporations and developers collaborate to advance autonomous driving software,” said Yang Zhang, chairman of the Autoware Foundation’s board of directors. “Our collaboration with TIER IV strengthens the international framework for validating and refining E2E autonomous driving through real-world deployment. By testing across three continents, we are driving standards-based innovation and expanding an open ecosystem that lowers the barrier for a diverse range of partners to join and contribute.”
“The release of these software stacks and MLOps platform is a vital step toward deploying advanced AI models in industrial applications,” said Yutaka Matsuo, professor at the University of Tokyo, Graduate School of Engineering. “By accumulating data from Japan’s distinctive traffic environments through our Tokyo testing and contributing those insights back to Autoware, we aim to further bridge the gap between academic research and real-world deployment.”
“Autoware is a foundational technology for shaping the Level 4+ autonomy concept,” said Raj Rajkumar, George Westinghouse Professor in the Department of Electrical and Computer Engineering at Carnegie Mellon University. “Our Pittsburgh testing will validate the effectiveness of this technology under unique urban traffic conditions. It is essential for the global advancement of autonomous driving that academia and industry continue to collaborate and share results through the Autoware ecosystem.”
“This initiative provides a valuable opportunity to evaluate technologies at the Level 4 autonomous driving standard within European urban environments and verify their effectiveness from multiple perspectives,” said Johannes Betz, Professor of Autonomous Vehicle Systems at the Technical University of Munich. “We expect that this framework—improving AI models using region-specific datasets through Autoware-based collaboration—will significantly contribute to the development of highly practical autonomous technology.”
*Autoware is a registered trademark of the Autoware Foundation.