AI Automated Translation.

Font Size

Share

"No need for 20-million-won GPUs"... This place is driving industrial innovation with cost-effective AI

"No need for 20-million-won GPUs"... This place is driving industrial innovation with cost-effective AI

[Interview] Lee Hoon of AIM T1

[For more diverse corporate information on the startups mentioned in this article, please visit Unicorn Factory's big data platform 'Data Lab'.]

=>AIM T1
=>AIM T1

"In the future, the competitiveness of industrial AI will not be determined by how large a model is created, but by how deeply it understands the actual work site."

Lee Hoon of AIM T1 recently stated in an interview with Money Today's startup media platform Unicorn Factory that "if general-purpose AI is a model that knows everything, vertical AI is an expert who knows a specific industry best," adding, "Ultimately, at industrial sites, AI that understands the on-site context will create greater value than general-purpose AI."

AIM T1 originated as an in-house venture of Hyundai Motor in 2000. It grew by developing an AI safety driving solution that detects over 30 types of risky behaviors in real time using vehicle interior and exterior camera footage, such as drowsy driving, mobile phone use while driving, failure to pay attention to the road ahead, and running red lights. Based on driving data, it calculates individual driver safety scores and provides customized coaching and vehicle management services, contributing to accident prevention for companies and reductions in insurance premiums and operating costs.

AIM T1 /Photo=AIM T1
AIM T1 /Photo=AIM T1

From hardware subcontracting to AI-driven transformation

The company is moving away from its car-centric business model to develop its own 'Vision AI' models, expanding its business scope into manufacturing, mobility, healthcare, and security. Lee said, "In the past, we developed lane departure prevention and color recognition technologies for the first time in the world, but growth was limited within the subcontracting structure of large corporations. We have now shifted our business direction to integrate existing technologies with proprietary AI models to secure independent competitiveness."

Currently, it supplies Vision AI inspection solutions in the manufacturing sector, facial recognition access control systems in security, and enterprise AI chatbots in customer support. In healthcare, it is also developing an AI solution that analyzes electrocardiograms (ECG) to diagnose arrhythmias.

At the core of these solutions lies its proprietary AI model, 'aimNet'. Its key feature is optimization to run AI smoothly even on low-cost semiconductors from Ambarella and Rockchip in the 20,000-won range, rather than relying on NVIDIA GPUs costing over 20 million won.

What AIM T1 cites as the source of its competitiveness is data accumulated at actual work sites. The company became the first in Korea to receive Class A certification for unstructured data quality from an institutional investor under the Ministry of Science and ICT. Lee explained, "We split over 11 million driving and accident videos collected from approximately 10,000 pieces of equipment into 1-second or 0.5-second segments to analyze context before and after each segment." He added, "The core asset is the technology and know-how to accumulate on-site data and use it to improve AI accuracy."

Aiming for the 'last 10%' of manufacturing sites

The area AIM T1 is currently focusing on is smart factories. Lee stated, "China has already implemented 'dark factories' with only one person per ten production lines operating without lights," and predicted that "within three to four years, a wave of automation will spread even to first- and second-tier suppliers in Korea."

AIM T1 utilizes the production lines of its largest shareholder, Dreamtek, and its affiliate Namuga as testbeds to repeatedly verify AI models and accumulate manufacturing data.

In particular, the area the company is targeting is the 'last 10%' of manufacturing sites. While most production lines are automated, final appearance inspections still often require humans to directly check for scratches, missing parts, or assembly defects.

AIM T1 explained that by utilizing AI vision technology, it can detect micro-defects down to the 3rd-micrometer (㎛) level and has achieved performance with a miss rate of 0% and an over-inspection rate of 3% or less. The company is also building a system that links finished products inspected for appearance to AMRs (autonomous mobile robots) for automatic transfer to the next process.

Lee noted, "Large corporation production lines are already highly automated, leaving few new areas for AI to enter." He added, "On the contrary, first- and second-tier suppliers with insufficient in-house AI development capabilities represent a much larger market and potential customer base." He further stated, "Ultimately, labor costs are the last element of cost reduction at manufacturing sites," adding, "Our goal is to reduce labor costs and increase productivity by automating part transport and final appearance inspections."

=>[Money Today startup media platform Unicorn Factory]

"This article was translated using AI and may differ slightly from the original."