Tier IV will showcase an integrated software-defined vehicle solution at Automotive World 2026 at Makuhari Messe in Tokyo from September 9 to 11. The demonstration covers the full development pipeline from data collection and processing through to the deployment of an end-to-end autonomous driving AI model on an automotive computing platform, running on an NVIDIA Jetson Orin.

Central to the demonstration is the autolabelling function of Tier IV’s Co-MLOps platform, which automatically labels vehicles, pedestrians, road surfaces, and structures in driving data to generate training datasets for autonomous driving AI. The company says the function can generate millions of labels instantly at consistent quality, replacing manual labelling work and reducing development costs. Rare or difficult-to-collect scenarios — including adverse weather and collision-risk situations — are supplemented using synthetic data generated via NVIDIA Cosmos, which is then processed through the same autolabelling pipeline.

The reference end-to-end AI model developed using this data operates without high-definition maps, relying solely on automotive camera images to perform bird’s-eye-view environment understanding, 3D object recognition, occupancy mapping, and trajectory generation within a single neural network. Development of the model is led by an agentic AI system that autonomously runs cycles of training, evaluation, and improvement based on human-specified objectives, with the architecture designed for efficient deployment on automotive system-on-chips.

Tier IV launched the Co-MLOps platform in January 2024 and has been building out a partner ecosystem for driving data collection and sharing. The company says it plans to make the reference E2E model available to Co-MLOps project partners as development continues.

Source: PRNewsWire