AI Times recently highlighted A.I.Matics’ proprietary AI technology and industrial AI strategy. In the interview, CEO Hoon Lee discussed the company’s on device AI technology, validated through data collected from more than 520,000 vehicles, its Automated AI Training System (AATS), and its vision for expanding Vertical AI beyond mobility into manufacturing, security, and healthcare. He also outlined the company’s ambition to develop a World Model AI capable of understanding and learning from industrial environments autonomously, setting a new benchmark for the Physical AI era.
[Interview]
As Vision AI rapidly reshapes industries, one company is demonstrating strong real-world results through proprietary AI models designed to run directly on edge devices. Founded as an internal venture within Hyundai Motor Group, A.I.Matics has built a strong technological foundation in mobility and is now rapidly expanding into smart manufacturing, security, healthcare, and other industries.
Built for On Device AI from the Ground Up: “Real World Data Is the Key”
A.I.Matics’ core competitive advantage lies in aimNet, a proprietary AI model designed from the outset to operate entirely on devices without relying on external servers or cloud infrastructure.
“Industrial environments where safety is directly linked to human lives cannot tolerate even a single second of response delay,” said Hoon Lee, CEO of A.I.Matics. “We developed our models specifically for on device deployment from day one, enabling real time decision making even when connectivity is unavailable while ensuring data security and reliability.”
According to Lee, the company’s proprietary AI is built upon extensive datasets accumulated through years of real-world deployments. Leveraging experience gained from supplying solutions to automotive OEMs, A.I.Matics has refined its models using an autonomous driving dataset consisting of more than 11 million video clips totaling 34 terabytes of data.
“This real world data asset, equivalent to approximately 3,500 years of accumulated driving footage, represents a practical competitive advantage that is extremely difficult for new entrants to replicate,” Lee emphasized.
주요 오픈소스 모델(Yolo) 대비 'aimNet‘ 성능 그래프 (사진=에이아이매틱스)
[Interview]
As Vision AI rapidly reshapes industries, one company is demonstrating strong real-world results through proprietary AI models designed to run directly on edge devices. Founded as an internal venture within Hyundai Motor Group, A.I.Matics has built a strong technological foundation in mobility and is now rapidly expanding into smart manufacturing, security, healthcare, and other industries.
Built for On Device AI from the Ground Up: “Real World Data Is the Key”
A.I.Matics’ core competitive advantage lies in aimNet, a proprietary AI model designed from the outset to operate entirely on devices without relying on external servers or cloud infrastructure.
“Industrial environments where safety is directly linked to human lives cannot tolerate even a single second of response delay,” said Hoon Lee, CEO of A.I.Matics. “We developed our models specifically for on device deployment from day one, enabling real time decision making even when connectivity is unavailable while ensuring data security and reliability.”
According to Lee, the company’s proprietary AI is built upon extensive datasets accumulated through years of real-world deployments. Leveraging experience gained from supplying solutions to automotive OEMs, A.I.Matics has refined its models using an autonomous driving dataset consisting of more than 11 million video clips totaling 34 terabytes of data.
“This real world data asset, equivalent to approximately 3,500 years of accumulated driving footage, represents a practical competitive advantage that is extremely difficult for new entrants to replicate,” Lee emphasized.
Performance Comparison of aimNet vs. Leading Open Source Models (YOLO)
(Source: A.I.Matics)
The performance of the company’s AI models is also noteworthy. aimNet v3 is approximately 1 MB smaller than the average open source model, which typically measures around 18 MB, yet delivers 50% higher recognition accuracy and up to 22 times faster processing speeds, enabling stable operation even in resource constrained environments.
To further accelerate development, A.I.Matics has built AATS (Automated AI Training System), which automates the entire AI lifecycle—from data collection and model training to deployment and continuous optimization.
Beyond Object Detection: Seven Core Cognitive Capabilities in a Single Video Stream
A.I.Matics combines its proprietary AI models with Vision Language Model (VLM)-based AI agents and a multimodal architecture.
This enables a single video stream to simultaneously perform seven key perception tasks:
- Object detection
- Driver Monitoring System (DMS)
- Tracking
- Keypoint detection
- Facial recognition
- Incident detection
- Depth estimation
This enables a single video stream to simultaneously perform seven key perception tasks:
Seven Core Perception Capabilities Powered by a Multimodal AI Architecture
(Image: A.I.Matics)
Seven Core Perception Capabilities Powered by a Multimodal AI Architecture
(Image: A.I.Matics)
“Traditional vision systems remain limited to recognizing objects based on predefined rules created by humans,” Lee explained. “Our AI learns relationships within data, allowing it not only to understand situations in real time but also to proactively detect risks and predict what may happen next.”
For applications requiring more sophisticated reasoning, A.I.Matics employs an edge cloud hybrid architecture, connecting edge devices to larger AI models to maximize computational efficiency while maintaining system scalability and reliability.
These technologies have already been deployed in more than 520,000 vehicles, supporting organizations including public transportation operators, Hyundai Motor Group, Kia, Samsung Electronics, LG Electronics, and Sampyo.
Expanding Beyond Mobility: Building a World Model for Industrial Environments
A.I.Matics is now extending the capabilities proven in mobility to a broader range of industries.
In particular, the company’s Vision AI inspection solutions for smart factories have demonstrated strong performance in identifying complex defect patterns that conventional rule based inspection systems struggle to detect.
Looking ahead, the company’s ultimate goal is to develop a comprehensive World Model capable of understanding and responding autonomously to entire industrial environments.
“Our objective is to build A.I.Matics’ own industrial World Model by applying advanced AI methodologies such as few shot learning and in context learning,” said Lee. “By extending these capabilities across industries, we aim to establish a new standard for the Physical AI era.”
Source: AI Times (Original interview published in Korean)
