Solving Field Problems with AI: 2026 AI Expert Course

As AI technology rapidly advances, how to solve real-world industry challenges using cutting-edge AI technology has emerged as a core element of corporate competitiveness.

 

'AI Expert Course,' organized by LG AI Research's Academy Team, is an intensive training program designed to foster practical AI experts. In this program, domain specialists from LG affiliates collaborate one-on-one with expert mentors from LG AI Research to define problems, leverage state-of-the-art AI technologies, and derive actionable solutions for field deployment. Over a 12-week period (April 7 – June 25), participants stay on-site for three days a week to focus on solving their challenges and maximizing practical applicability.

 

Image 1. A Glimpse of the LG AI Expert Course 2026 Achievements Sharing Session

 

On September 4, '2026 AI Expert Course Achievement Sharing Session' was held at Magok K-Square to mark the conclusion of this 12-week journey. Domain specialists—including representatives from LG Display and LG Electronics—along with members of LG AI Research's Academy Team and expert mentors, attended the event to share the achievements of tailored, field-ready AI solutions.

AI Technology Application Achievements by Industry Sector

During this course, a total of 10 field-relevant projects were conducted across three main domain areas addressing complex operational challenges:

• • Data Intelligence (DI) Domain (3 projects): Focused on enhancing operational efficiency by abstracting, interpreting, predicting, and optimizing complex data in manufacturing and process management.
    

• • Language Domain (3 projects): Centered on deep learning-based LLMs, document understanding, and STT/TTS technologies.

• • Physical Intelligence (PI) Domain (4 projects): Aimed at realizing physical spatial intelligence, covering computer vision and robotics.

At this sharing session, three projects from the DI and PI domains were selected as top-performing projects for demonstrating outstanding research results and high applicability to field operations. These cases proved that the capabilities of LG’s AI talent can materialize into customized, on-site AI solutions while expanding virtual intelligence into real-world technological innovations.

1st Place: On-Device AI Scalable Across All Product Lines Starting from Core Data

Image 2. Sehwa Choe from LG Electronics

Sehwa Choe, Senior Research Engineer, LG Electronics (Mentor: Minkook Suh, DI Lab)

Training AI for air conditioner fault diagnosis typically requires collecting large volumes of failure data across various product lines, which incurs significant time and cost. This project developed a methodology to filter critical fault scenarios and complement insufficient data using advanced AI techniques. As a result, identical diagnostic performance was achieved with only 30% of the original dataset, laying the groundwork to significantly reduce future development lead times and costs.

 

Sehwa Choe noted, "Through the 1-on-1 mentoring, I was able to view complex field issues from a theoretical perspective. My mentor kindly and meticulously guided me whenever I hit a wall, giving me the confidence to grow further."

[Comment from Mentor Minkook Suh, DI Lab]

"By exploring various approaches such as Domain Adaptation and Evolutionary algorithms together with Sehwa, we were able to tackle challenging technical hurdles. The data synthesis technology developed in this project holds immense potential for expansion into other domains. I sincerely thank Sehwa for showing such passion throughout the program, and congratulations on winning 1st place!"

2nd Place: Pushing the Limits of Humanoid Motion Control – MoE-based Universal Whole-Body Control

Image 2. Yunsoo Kim from LG Electronics

Yunsoo Kim, Senior Research Engineer, LG Electronics (Mentor: Whiyoung Jung, PI Lab)

Enabling humanoid robots to stably execute diverse action commands under limited computing resources is a highly challenging task. This project applied adaptive motion resampling to encourage more frequent learning of difficult motions, thereby reducing bias, and restructured a single MLP architecture into a Mixture of Experts (MoE) model to enhance expressiveness. Through this, a reinforcement learning model capable of expressing significantly more actions with a comparable parameter count was secured.

 

Yunsoo Kim shared, "I had many concerns in the fast-moving robotics sector, but by applying the Physical AI knowledge and problem-solving methodologies refined with my mentor, I was able to deliver meaningful insights to the entire control team at LG Electronics."

 

[Comment from Mentor Whiyoung Jung, PI Lab]

"Yunsoo successfully led demanding tasks ranging from motion data cleaning to reinforcement learning reward design. Thanks to his efforts, achieving 2nd place is a wonderful outcome. Exceptional work!"

3rd Place: Rule vs AI – Optimization of Robot Transfer Operations in Manufacturing Processes

Image 3. Youngbeen Moon from LG Display

Youngbeen Moon, Senior Research Engineer, LG Display (Mentor: Junseok Park, DI Lab)

Factory robot operation strategies directly impact overall production output. While conventional methods relied on manually exploring cases via simulation, this project applied reinforcement learning to develop an AI model that determines optimal robot actions based on real-time equipment status. Ultimately, a production yield increase of up to 0.9% compared to traditional rule-based operations was validated.

 

Youngbeen Moon expressed his gratitude, saying, "By following mentor Junseok Park’s guidance, I learned an immense amount and achieved a great result in 3rd place. I extend my sincere thanks to my mentor for his dedicated support."

[Comment from Mentor Junseok Park, DI Lab]

"Building a simulation environment from scratch and implementing reinforcement learning within a short timeframe made setting priorities our biggest challenge. We stabilized the learning algorithm and environment first before refining the reward function, which enabled us to achieve performance surpassing existing field heuristics. I am extremely proud of Youngbeen for tackling complex industrial challenges with reinforcement learning and achieving such a meaningful result."

The Future of the AI Expert Course

Image 4. Group Photo of the LG AI Expert Course 2026

The 2026 AI Expert Course Achievement Sharing Session concluded on a warm note as mentors and mentees shared their research outcomes and exchanged congratulations and encouragement.

 

LG AI Research plans to continuously advance the AI Expert Course to support the seamless integration of its advanced AI capabilities into various LG affiliate sites. We look forward to seeing more 'One-Team' AI talents who independently define field problems and solve them with cutting-edge AI technologies.