Collaborative data and AI platform
Accelerating AD/ADAS development
The Co-MLOps platform has been designed to scale the development of autonomous driving AI.
About Co-MLOps
The more participants that join, the greater the benefit for all
Developed by TIER IV, the Co-MLOps platform makes it possible to accelerate AI development using large-scale shared data collected from around the world.
Data Pipeline Workflow
Enhance autonomous driving AI development
Sensor data such as camera images and LiDAR point clouds collected from various regions around the world can be shared through the Co-MLOps platform. With MLOps features and edge AI reference models, the platform equips companies to enhance autonomous driving AI.
Large-scale Data Collection
Featuring a reference
sensor suite
To support the rapid development and optimization of autonomous driving systems, we offer comprehensive services including advanced data collection, cloud management, and image anonymization for privacy protection.
-
Data collection vehicles are equipped with time-synchronized and calibrated sensors such as cameras and LiDAR, enabling high-precision recording of road conditions and surrounding environments.
-
Collected data is securely stored and efficiently managed in a cloud-based system. The platform supports seamless data retrieval, with automated and manual annotation features.
Protect privacy and secure data
Anonymization Technology
Privacy is safeguarded by anonymizing license plates and human faces. Vast amounts of data are collected from vehicles during autonomous driving. Protecting privacy is crucial. On the cloud side, data anonymization ensures secure data management, while on the edge side, real-time data processing and anonymization at the point of collection reduce privacy risks.
AI-generated Face Replacement
Face Blurring
Active Learning Framework
Active learning features for Al training
Using the collected dataset, uncertainty and reliability are quantified. Specifically, the model's confidence in its predictions is assessed, and areas with high uncertainty are identified to improve the system's accuracy and reliability.
Active learning features optimize data selection for training, even with limited datasets. Labeling and training in areas with high uncertainty are prioritized, among other factors.
Reference AI
Full surround perception for AD/ADAS
The reference AI model is a lightweight model that supports common recognition functions used in AD/ADAS applications. It serves as a foundational model upon which developers can build their own algorithms, facilitating the efficient development of advanced autonomous driving systems.
Data Recording System (DRS)
A robust suite of sensors and computing systems for advanced data collection
The latest-generation automotive sensors enable high-precision 360-degree recording of the vehicle's surroundings over long distances. All devices are synchronized using PTP (Precision Time Protocol) or FSYNC, ensuring seamless operation. The system includes calibration for intrinsic camera parameters, as well as extrinsic parameters between multiple cameras and LiDAR sensors. Equipped with an included storage system, it supports extended data collection sessions. With this DRS (Data Recording System), you can achieve high-quality data collection essential for autonomous driving development.
Annotation samples
-
![]()
Panoptic segmentation
Panoptic Segmentation is a technique that assigns pixel-level labels to every object in an image. It distinguishes not only the shapes of objects but also the background and different instances of the same type of object. This enables a more detailed and complex understanding of scenes compared to other annotation methods, allowing it to handle a wider range of tasks.
-
![]()
Traffic signal recognition
The platform supports the development of various traffic light recognition systems, with traffic signals enclosed in bounding boxes and each signal light labeled with attributes such as color, type, and arrow direction.
-
![]()
3D bounding box
Based on the 3D point cloud data collected by LiDAR, a bounding box is defined in the three-dimensional space to encompass the position and shape of an object. This box is used for object detection and tracking.
Partnerships
-
Hitachi Astemo
Hitachi Astemo has been involved in Co-MLOps since the 2023 launch of the pilot program, which included data collection across eight global locations. As a lead partner since 2024, Hitachi Astemo has played a key role in advancing the platform’s functionalities, focusing on autonomous driving platforms and software-defined vehicles.
-
Nihon Kotsu
Since July 2024, taxi operator Nihon Kotsu and TIER IV have been collaborating to promote the development of a large-scale shared data infrastructure. By utilizing vehicles equipped with TIER IV’s Data Recording System, we are jointly building extensive and diverse datasets, contributing to the development of safe autonomous driving technology.
-
The Autoware Foundation
Established in 2018, the Autoware Foundation oversees the development of Autoware, the world’s first open-source software for autonomous driving, with TIER IV leading efforts to advance the technology while nurturing a collaborative community.
