[AI Brief] Konan Technology Launches Gyeongnam Physical AI State Project

Konan Technology said on the 18th that it attended the Ministry of Science and ICT-led launch briefing for the Jeonbuk Special Self-Governing Province and Gyeongnam Physical AI R&D project on the 16th.

At the briefing, Konan Technology shared the project direction and unveiled its R&D roadmap.

The Gyeongnam project focuses on developing a human-AI collaboration-type Large Action Model (LAM). The project is a major R&D initiative aimed at AI that goes beyond large language models (LLMs) and carries out real-world actions.

The project will move forward by combining AI with physical equipment such as robots and machines and by training on data from manufacturing sites.

Konan Technology said it aims to realize autonomous control of facilities and robots based on training data, and that it will also pursue global demonstrations under the project.

Starting this year, the company will build an integrated platform and verification system related to manufacturing process data, data pipelines, and bidirectional digital twins. To that end, it will invest a total of KRW 676.3 billion over 5 years through 2030. The purpose of the investment is to pursue LAM development and phased demonstrations at manufacturing sites.

Konan Technology, Megazone, and the Electronics and Telecommunications Research Institute (ETRI) formed a consortium. The consortium will participate in the overarching task, 'Development of Technologies for Securing the Reliability of Manufacturing Data and an Integrated Lifecycle Operations Platform.'

Konan Technology is responsible for applying agent technology. Specifically, it will develop technology to automatically convert unstructured manufacturing-site data into AAS format. It will also work on quality management technology for measuring and correcting the gap between digital twin synthetic data and real-world physical data.

Meanwhile, Xenon said on the 18th that it will release the AI model 'Hunmin VLM 397B' as open source. The photo credit is Xenon.

'Hunmin VLM 397B' has the ability to perform tasks through computer screen recognition and direct manipulation. The model is based on 'Qwen(Qwen3.5-397B).' 'Qwen(Qwen3.5-397B)' has 397 billion parameters. Xenon said it strengthened 'Hunmin VLM 397B's' ability to recognize computer screens and carry out direct manipulation.

Hunmin VLM 397B is a model that goes beyond question answering and information generation to understand a user's work environment and act directly. Its core lies in improving both the ability to find targets for manipulation on the screen and the ability to carry out actual computer tasks. Hunmin VLM 397B has improved grounding performance for detecting targets such as buttons and input fields on the screen, and its overall computer operation performance, including task execution in real computer environments, has also improved.

Xenon said Hunmin VLM 397B outperformed both the base model Qwen3.5-397B and the computer-operation-specialized model Qwen-CUA on 5 benchmarks that identify the location of objects to be manipulated on the screen. Xenon said this shows the model improved performance across both target detection and actual computer manipulation.

The model applies Xenon's post-training technology, which upgrades a 397B-scale large open model with limited computing resources. Xenon transferred the capabilities of an open computer-operation model using a low-rank method and carried out supervised fine-tuning (SFT) and reinforcement learning (RL) in FP8 quantized form. The resources used for the entire post-training process were 8 NVIDIA B200 graphics processing units (GPUs).

Freewheelin is moving to establish a system that applies AI to university education after signing a business agreement with Sejong University. Freewheelin said on the 18th that it signed an AI-based teaching and learning innovation agreement with Sejong University. The purpose of the agreement is to drive AI-based teaching and learning innovation and strengthen university education competitiveness.

Freewheelin will work with Sejong University to promote AI-based teaching and learning innovation. The two institutions plan to build an educational system that links first-year students' basic academic diagnostics, personalized supplementary learning, and AI writing evaluation using the university education-focused AI courseware 'Pulli Campus.' The scope of the joint build covers an educational system that includes first-year students' basic academic diagnostics, individualized supplementary learning, and AI-based writing evaluation.

Specifically, the two institutions will operate adaptive basic academic diagnostic tests for first-year students and run customized supplementary learning for first-year students. They will also pilot AI-based writing evaluation. In addition, they will use pre-learning content for introductory major courses and course assessments for introductory major courses.

The two institutions plan to link Sejong University's LMS 'Jiphyeon Campus' with Pulli Campus accounts. They will also conduct performance analysis based on operational data and consider expanding adoption.

In the photo provided by Freewheelin, Sejong University President Um Jong-hwa and Freewheelin CEO Kwon Gi-seong are shown. The photo source is Freewheelin.

The two institutions plan to run a pilot program in the second semester. During the pilot, they plan to collect usage rates. During the pilot, they also plan to collect satisfaction levels.

During the pilot, the two institutions plan to collect learning data. They plan to evaluate operating performance in December.

Based on the evaluation results, the two institutions plan to review a plan to expand the basic academic diagnostics to all students. Based on the evaluation results, they also plan to gradually review ways to expand the areas of use.

The areas under consideration for expansion are major education, career and job placement education, and Korean language education (TOPIK) for international students.

Meanwhile, Wonderful is expanding the scope of enterprise AI adoption to the sports industry. Wonderful plans to apply AI agents across club operations and business activities. On the 18th, Wonderful announced a 3-year global partnership with Bundesliga soccer club FC Bayern Munich of Germany.

Wonderful will serve as FC Bayern's 'official enterprise AI partner.' Wonderful plans to support AI adoption in real-world settings across the club's operations. It will introduce enterprise AI solutions to fan services and other major use cases.

As its first use case, the two companies introduced an AI agent specialized in supporting 'matchday fan service.' The AI agent is dedicated to digital ticket support on game days, helping handle inquiries related to ticket access and accounts and enabling immediate resolution of on-site fan inquiries without on-site waiting lines. During the partnership, the two companies also plan to sequentially unveil additional enterprise AI use cases for advancing club operations and business activities.

Source: IT DAILY · Yang Seung-gap
Original: https://www.itdaily.kr/news/articleView.html?idxno=241710

References

• https://www.itdaily.kr/news/articleView.html?idxno=241710