The demonstration, running from September 9 to 11, showcases a workflow built on the Co-MLOps platform. TIER IV aims to solve the bottleneck of manual data preparation by introducing an autolabeling function capable of generating millions of labels instantly. This system identifies vehicles, pedestrians, and road structures, ensuring consistent data quality even as sensor configurations evolve or data volume scales. To address rare collision risks or adverse weather, the platform integrates synthetic data generated through NVIDIA Cosmos, further refining the training sets.

The centerpiece of the exhibit is a reference end-to-end AI model that moves away from traditional high-definition maps. By relying exclusively on automotive camera feeds, the model uses a single neural network to handle complex tasks, including bird's-eye-view environmental mapping, obstacle detection, and trajectory prediction. Developed through an agentic AI process, the model automates the cycle of training, evaluation, and optimization to accelerate prototyping cycles.

Running on an NVIDIA Jetson Orin platform, the demonstration proves that this sophisticated software can operate efficiently on automotive-grade hardware. By providing this integrated architecture, TIER IV intends to assist automakers and suppliers in simplifying the transition toward mass-market autonomous passenger vehicles.