#Japan #TIER IV #Automotive World #Shinagawa #SDV Solutions

Introduction

Tier IV, a company dedicated to the democratization of autonomous driving, will showcase its innovative solutions at Automotive World 2026, held from September 9-11 at Makuhari Messe. With a focus on Software Defined Vehicles (SDVs), the company aims to demonstrate how AI, data, and computing can be seamlessly integrated to enhance autonomous driving capabilities.

Exhibition Highlights

One of the primary features at Tier IV's booth is the Co-MLOps Platform, which includes an automatic labeling function. This platform is designed to facilitate the development of end-to-end (E2E) autonomous driving AI models, specifically showcasing the Reference E2E AI Model. Attendees can expect live demonstrations of how the model operates on an in-car computer, showcasing an all-encompassing solution that blends data collection, processing, and execution for SDV applications.

Importance of High-Quality Data

As the adoption of AI in autonomous driving accelerates, the quality and diversity of labeled data become crucial to the performance of AI systems. Tier IV recognizes that the efficiency of deploying autonomous driving AI models to mass-produced vehicles is contingent upon the effective operation of these models on in-car computing systems. Thus, they have made it a priority to ensure that high-quality data underpins the development of their autonomous driving solutions.

Co-MLOps Platform

Scheduled to launch in January 2024, the Co-MLOps Platform aims to expedite the pace of autonomous driving AI development. Tier IV is working with partner companies to gather and share driving data, which will then be automatically labeled through the platform. This approach not only streamlines the data labeling process but also fosters the efficient development of E2E autonomous driving AI models while establishing the necessary environment for execution on in-car computing systems.

Features of Automated Labeling

To train autonomous driving AI models effectively, a significant volume of driving data is required, alongside accurate labeling for correct interpretation. The automated labeling feature of Tier IV’s platform allows for the precise identification of various objects and backgrounds present in driving data, such as vehicles, pedestrians, road surfaces, and infrastructure. The platform can generate millions of labels at a consistent quality instantaneously, ensuring efficient data processing, even as new regions, sensors, or data volumes are brought into the equation.

Enhancing Model Accuracy

Moreover, to overcome the challenges of acquiring rare driving scenarios linked to collision risks—such as uncommon weather conditions—Tier IV supplements its collected driving data with generated data from NVIDIA's Cosmos. Even this generated data undergoes automatic labeling, further improving the robustness of the overall model.

For an in-depth understanding of the automated labeling capabilities and how NVIDIA Cosmos enriches the data dataset for autonomous driving, readers can explore Tier IV’s tech blog titled "Building Autonomous Driving Data Sets with NVIDIA Cosmos."

Features of Reference E2E AI Model

The Reference E2E AI Model operates without relying on high-precision maps, harnessing only the footage from onboard cameras. It is designed to recognize its environment, generate paths and maneuver around obstacles using a singular neural network. This architecture simplifies the integration into actual system-on-chip (SoC) units, promoting ease in practical applications.

In developing this model, a unique collaboration between human directives and agent-based AI takes place. The AI leads the development process, tackling the ingestion of automatically labeled data, data transformations, cleansing, optimization, and model management. This symbiosis allows for an agile and cost-effective prototyping cycle, dramatically improving development efficiency.

Future Plans

Tier IV intends to continue the development of the Reference E2E AI Model and extend its platform as a vital resource for companies participating in the Co-MLOps Project, which focuses on extensive data collection and sharing.

Demonstrations at Automotive World 2026

During this year’s Automotive World event, Tier IV will provide demonstrations showcasing the automatic labeling technology through their Co-MLOps Platform. Attendees can also observe how the labeled driving data contributes to the functionality of the Reference E2E AI Model, executed on NVIDIA Jetson Orin, enabling environment recognition and path generation using only camera input.

As Tier IV integrates large-scale driving data collection and processing, continuous development, and enhancement of E2E autonomous driving AI models and their implementation on in-vehicle computing, they are paving the way for the advancement of SDV systems, ensuring that automotive manufacturers can effectively develop autonomous systems for consumer vehicles.

About Tier IV

Founded in December 2015 and headquartered in Shinagawa, Tokyo, Tier IV is at the forefront of the open-source autonomous driving software movement with their development of Autoware. As a deep tech company, they offer a comprehensive software platform based on Autoware, tailoring services according to market needs. Through partnerships across the globe, Tier IV strives to enhance the ecosystem that Autoware creates, aiming to expand the potential of autonomous driving while working towards a safer and more sustainable society.