Kaist

˝Age of Robots Making Human-Like Judgments, KAIST ..

< (From left) Professor Chang D. Yoo, Tung M. Luu (PhD candidate, first author) at the back center, and Hwanhee Kim (M.S candidate, second author) at the front right > “Robots that make judgments like humans are coming faster than we think.” A core technology that will accelerate the era where robots understand human intentions and choose the correct actions on their own has been developed in South Korea. KAIST researchers solved a key challenge in the commercialization of physical AI by developing a technology where AI learns human judgment criteria on its own with just a few videos. KAIST announced on June 10th that a research team led by Professor Chang D. Yoo from the School of Electrical Engineering has developed 'VOTP (Video-based Optimal TransPort Preference)' for the first time in the world, a new technology that allows AI to learn human intentions and judgment criteria using just a few preference videos instead of thousands to tens of thousands of human evaluation data points. < VOTP Overview Diagram > The research team's paper has been accepted to ICML (International Conference on Machine Learning) 2026, the world's most prestigious AI conference, which will be held at COEX in Seoul this July. It was selected for an Oral presentation, an honor given to only the top 0.7％ (168 papers) out of all submitted papers (23,918 papers), recognizing the excellence of the research. ICML is considered one of the most influential international conferences in the fields of AI and machine learning. Recently, AI technology is rapidly evolving beyond generative AI that writes text and draws pictures into the era of 'Physical AI,' which moves actual machines and acts in the real world. Representative examples include robots that perform dangerous tasks in factories instead of humans, autonomous vehicles that judge road situations on their own, and medical robots that perform delicate surgeries. However, there was a barrier that had to be overcome for the practical application of physical AI. It is the problem of learning human-level evaluation criteria to judge whether the actions performed by a machine match human intentions and which actions are more desirable. For example, when a surgical robot performs suturing or an autonomous vehicle passes through a complex intersection, the AI must choose the most appropriate action among numerous options. To achieve this, a 'Reward Function' that reflects human preferences and judgment criteria is required. However, until now, humans had to directly evaluate thousands to tens of thousands of action data points to build this, which required an enormous amount of time and cost. The research team focused on the way humans learn new tasks after seeing just a few demonstrations. VOTP, developed by the research team, helps AI understand human-preferred action patterns on its own with just a few videos of good and bad examples. Even without humans evaluating a vast amount of data one by one as before, AI can understand human judgment criteria and expand its learning to various situations. The core idea of this research is that intelligent machines such as robots or autonomous vehicles can quickly grasp human intents with only a small number of videos containing human preferences. The algorithm developed for this purpose proved its effectiveness and generalization performance through extensive experiments across various environments and tasks. This method can significantly reduce human feedback and data construction costs required for physical AI development. Since robots, autonomous vehicles, and industrial machinery can learn actions that meet human expectations with only a small number of examples, it is expected to drastically shorten development time and costs. The technology can be widely applied not only to robot arm control, humanoid robots, autonomous vehicles, smart factories, drones, and surgical robots, but also to AI agents that directly operate computers. In particular, it is expected to be utilized as a core foundational technology for all physical AI systems that need to learn human intention and satisfaction. < VOTP Research Image (AI Generated) > Professor Chang D. Yoo said, "The core of physical AI is making machines understand human intentions and choose the correct actions," and added, "Since VOTP can learn human judgment criteria with only a small number of videos, it is a core technology that will accelerate the era of robots making human-like judgments." This research, in which PhD student Tung M. Luu from the School of Electrical Engineering participated as the first author, was selected as an Oral presentation paper at ICML (International Conference on Machine Learning) 2026, the world's most prestigious AI conference. ※ Paper Title: Video-Based Optimal Transport for Feedback-Efficient Offline Preference-Based Reinforcement Learning, Paper File: https://sanctusfactory.com/data/file/publications/202606091714078906.pdf This research was conducted with support from the Institute for Information & Communication Technology Planning & Evaluation (IITP) and the National Research Foundation of Korea (NRF), funded by the Ministry of Science and ICT."

KAIST Turns DNA from Genetic Information Carrier i..

<Professor Jimin Park(Center), Ph.D candidate Tae Kyoung Lee, Ph. D candidate Sang Yeon Oh> The fixed idea that DNA is only a molecule that stores genetic information is being challenged. KAIST researchers have developed a technology that controls the chemical environment around catalysts at the nanometer scale by designing DNA sequences, the arrangement of A, T, G, and C that make up genetic information. The team has presented a new catalyst platform that can improve hydrogen production efficiency and increase the yield of desired chemical products by designing DNA much like writing a computer program. KAIST (President Kwang Hyung Lee) announced on the 21st of May that a research team led by Professor Jimin Park of the Department of Chemical and Biomolecular Engineering has developed a core technology that precisely controls the microscopic chemical environment around catalysts by coating the surface of gold nanoparticles, ultrafine gold particles measuring 1–100 nm, with “single-stranded DNA,”a flexible DNA molecule composed of a single strand that can be designed with a desired length and structure and serves as a nano-coating material for controlling the reaction environment. In electrochemical reactions, which use electricity to drive chemical reactions and are used for hydrogen production or the manufacture of eco-friendly chemicals, performance is determined not only by the catalyst itself but also by the local reaction environment around the catalyst, such as acidity (pH) and ion distribution. However, conventional approaches have relied on special polymer coating materials, plastic-like materials made of long molecular chains, and have faced limitations in precisely designing internal structures at the nanometer scale. To solve this problem, the research team focused on “single-stranded DNA,” DNA composed of a single strand. DNA carries a negative charge, meaning it can influence the movement of surrounding ions, atoms or molecules with electric charge, and it has the advantage that its length and base sequence can be freely designed. In particular, changing the base sequence allows the internal network structure of DNA to be precisely controlled, making it possible to create a customized nano-coating layer on the catalyst surface. < Schematic illustration of catalytic interfacial microenvironment regulation using a single-stranded DNA layer > The research team attached DNA with various base sequences to the surface of gold nanoparticles and analyzed the electrochemical reactions. As a result, they found that the key factor determining catalyst performance was not simply the thickness of the coating layer, but the internal network structure formed according to the DNA base sequence. This means that even coating layers of the same thickness can create different pathways for the movement of ions needed for reactions, depending on how the internal DNA structure is organized. It is the same principle as traffic flow changing depending on how a road network is designed, even when the roads are the same width. The team also used real-time surface-enhanced Raman spectroscopy, a technology that uses lasers to analyze the chemical state of molecules in real time, to observe the reaction process. Through this, they directly confirmed that the DNA layer functions as an interfacial layer, a layer that performs a special function at the boundary where two materials meet, by regulating the movement of hydroxide ions (OH⁻) and changing the local pH around the catalyst. In simple terms, the DNA layer acts like a “traffic control center” around the catalyst, guiding the movement of ions. It helps some ions move more quickly while restricting the movement of others, thereby changing the reaction environment in the desired direction. By observing this process in real time, the researchers proved that DNA is not merely a protective film, but actively regulates the reaction environment. The team applied this technology to the hydrogen evolution reaction and the glycerol oxidation reaction, which converts glycerol, a byproduct of biodiesel production, into high-value chemicals. As a result, hydrogen production efficiency varied significantly depending on the DNA base sequence, and the selectivity, the proportion of a specific product formed, for glycerate, a material used in cosmetics and pharmaceuticals, also improved. This means that desired reaction outcomes can be achieved simply by adjusting the DNA sequence, without newly creating complex catalyst structures. < Research image (AI-generated image) > Professor Jimin Park said, “This study shows that DNA can be used not as a genetic information storage medium, but as a precise nanomaterial that controls electrochemical reactions,” adding, “By designing DNA sequences to control acidity and ion movement on catalyst surfaces, we expect this technology to be broadly applied across carbon-neutral technologies, including hydrogen production and biomass conversion.” This study was conducted with KAIST Department of Chemical and Biomolecular Engineering doctoral students Sang Yeon Oh and Tae Kyoung Lee as co-first authors, and Professor Jimin Park as the corresponding author. The research was published on May 5 in the internationally renowned Journal of the American Chemical Society. ※ Paper title: “Programmable Single-Stranded DNA Layers as Modulators of Nanoscale pH at Electrocatalytic Interfaces,” DOI: 10.1021/jacs.6c02995 ※ Author information: Sang Yeon Oh and Tae Kyoung Lee (KAIST, co-first authors), Jaeyeon Jun, Jinse Woo, Changho Lee, Yongha Kim (KAIST, co-authors), and Jimin Park (KAIST, corresponding author) This research was supported by the National Research Foundation of Korea’s Outstanding Young Scientist Program , Global Matching Program , and Young Researcher Infrastructure Support Program.

KAIST Produces Eco-Friendly Core Nylon Precursors ..

<(From Left) Dr. Da-Hee Ahn, Distinguished Professor Sang Yup Lee> Nylon is a representative plastic material used throughout our daily lives, from clothing to automobiles. However, most of its raw materials have been produced through petrochemical processes, resulting in large carbon emissions. KAIST researchers have developed a technology that can produce key nylon precursors in an eco-friendly way using microbes. KAIST (President Kwang Hyung Lee) announced on the 31st of May that a research team led by Distinguished Professor Sang Yup Lee of the Department of Chemical and Biomolecular Engineering has developed an Escherichia coli-based modular platform capable of producing three key monomers (basic molecular units that make up polymers) of “nylon 6,6” and “nylon 6” — adipic acid, hexamethylenediamine, and epsilon-caprolactam — from “glycerol (an eco-friendly bio-based byproduct generated during biodiesel production),” a renewable carbon source, using systems metabolic engineering (a technology that designs and optimizes microbial metabolic pathways to maximize the production of desired substances). “Nylon 6” is highly flexible and is used in clothing and films, while “nylon 6,6” has excellent strength and heat resistance and is used in automobiles and machinery parts. The numbers after the nylon name indicate the number of carbon atoms contained in the raw material molecules. The core of this study is that the biosynthetic pathway was divided into upstream and downstream modules, with E. coli strains assigned different roles. The upstream strain was designed to produce adipic acid from glycerol, while the downstream strain was designed to convert it into hexamethylenediamine or epsilon-caprolactam, respectively. Through this, the research team succeeded in producing adipic acid and hexamethylenediamine, the key raw materials of nylon 6,6, and epsilon-caprolactam, the key raw material of nylon 6, within a single integrated platform. To improve production efficiency, the researchers compared and validated various enzymes (proteins that promote chemical reactions in living organisms), including carboxylic acid reductases and transaminases, and applied the optimal combination, thereby improving hexamethylenediamine titer. In addition, in the epsilon-caprolactam production process, they designed a flexible-linker fusion enzyme that enhances reaction efficiency through efficient cofactor regeneration.In the upstream module, the team reconstructed the biosynthetic pathway (a series of reaction processes through which compounds are produced in living organisms) and improved the performance of key enzymes using artificial intelligence (AI), increasing production titer. As a result, they succeeded in producing adipic acid at a level of 6 grams per liter (g/L) in a fed-batch fermentation process. The research team also applied a “delayed inoculation” strategy (time-staggered co-culture), in which the second strain is introduced later after sufficient adipic acid has first been produced, rather than adding the two types of E. coli simultaneously. This is a method of sequentially introducing microbes with different roles at different times. When this strategy was applied to a fed-batch fermentation process (a fermentation method that increases productivity by supplying nutrients step by step), the team produced 230 milligrams per liter (mg/L) of hexamethylenediamine and 808 micrograms per liter (μg/L) of epsilon-caprolactam using only glycerol. Although the production amounts are not yet high, the research team explained that these results represent world-class performance among cases of direct production from glycerol. <Schematic Diagram> This technology is significant in that it presents the possibility of producing nylon raw materials, which have relied on petrochemical processes, through bio-based methods. The research team plans to further improve titer by combining AI-based enzyme design with additional systems metabolic engineering, and to expand the platform to produce various polymer raw materials (substances formed by the repeated bonding of multiple monomers). Distinguished Professor Sang Yup Lee stated, “This study is meaningful in that it presents a modular microbial platform capable of producing key monomers required for nylon 6 and nylon 6,6 production from renewable carbon sources,” adding, “We will continue to advance enzyme and metabolic flux engineering to improve titer and develop this into a core platform for sustainably producing various bio-based polymer raw materials.” The results of this study were published on May 4 in the Proceedings of the National Academy of Sciences (PNAS), with Dr. Da-Hee Ahn of the Department of Chemical and Biomolecular Engineering as the first author. ※ Paper title: “Metabolic engineering of Escherichia coli for the biosynthesis of nylon 6 and nylon 6,6 monomers” Authors: Sang Yup Lee (KAIST, corresponding author), Da-Hee Ahn (KAIST, first author), Tong Un Chae (KAIST, second author), total of 3 authors DOI: https://doi.org/10.1073/pnas.2535786123 This research was supported by the “Development of Platform Technologies of Microbial Cell Factories for the Next-Generation Biorefineries” project under the Petroleum Replacement Eco-Friendly Chemical Technology Development Program supported by the Ministry of Science and ICT, and by the “Development of Advanced Synthetic Biology Source Technologies for Leading the Biomanufacturing Industry” project under the Core Synthetic Biology Technology Development Program.

˝Development of ＇ADvisor＇, an AI that Predicts Ins..

<(Bottom from left) M.S candidate Gyurim Hwang, M.S candidate Yeongho Kim, Ph.D. candidate Kyungho Kim, Ph.D.candidate Jongha Lee, M.S candidate Yeonje Choi (Top from left) Undergraduate student Sejin Chung, Researcher Hongseok Lee, Researcher Myeong Ho Song, Ph.D. candidate Sunwoo Kim, M.S candidate Juyeon Kim, Professor Kijung Shin> Social media advertising usually requires running multiple ad drafts in practice before determining which ad is effective. Because of this, testing advertisements demands significant time and costs. Furthermore, the criteria for an effective advertisement vary greatly by brand. While some brands prefer person-centered advertisements, others receive better responses from advertisements that emphasize actual usage scenes. However, these effective advertising strategies for each brand are often not clearly defined in the field, which has limited the technology to systematically reflect them and predict advertising performance. To solve this problem, a research team led by Professor Kijung Shin at KAIST, in collaboration with the AI marketing company MADUP, developed 'ADvisor', an AI technology that predicts advertising performance for each brand. ADvisor utilizes a generative vision-language model that understands images and text simultaneously to find different advertising success criteria for each brand and predict advertising effectiveness based on them. To achieve this, it not only analyzes the characteristics of the brand but also considers advertising data from other brands with similar tendencies for new brands that do not have sufficient advertising data to derive advertising strategies. Through this process, it can identify distinct advertising success criteria for each brand; for instance, a "strong headline phrase" is analyzed as an important criterion for a specific fashion brand, while "logo exposure" acts as a key element for another brand. Afterward, ADvisor evaluates the advertisement based on the derived criteria for each brand, reviews the evaluation results on its own, and repeatedly compensates for deficiencies to make the final prediction. The research team verified the technology's performance using data from 10 brands in the beauty, fashion, and platform sectors collected through actual marketing campaigns. As a result, ADvisor recorded up to 7.2％ higher performance compared to existing AI advertising prediction models. In particular, in an online A/B test conducted in a real Instagram advertising environment, it achieved an average of 27％ better performance in key indicators such as click-through rate (CTR), cost per click (CPC), and return on ad spend (ROAS) than advertisements selected by field marketing experts, proving that it can be utilized in actual marketing decision-making. Professor Kijung Shin stated, "Predicting advertising performance in advance is the first step for effective advertisement production," adding, "In the future, we will develop our research in a direction where AI directly generates and optimizes advertisements tailored to brand characteristics." The study, in which Ph.D candidate Kyungho Kim and M.S candidate Yeonje Choi from the KAIST Kim Jaechul Graduate School of AI participated as co-first authors, was published online on April 18 in the Industry Track of ACL 2026, one of the most prestigious international academic conferences in the field of natural language processing. It has been accepted as an oral presentation paper and is scheduled to be presented in the United States this coming July. ※ Paper Title: Pre-Deployment Advertisement Ranking under Data Scarcity via Context-Aware Criteria Generation with VLMs ※ Paper Link: https://openreview.net/forum?id=il84gAzAxx Meanwhile, this research is an achievement of the project 'EntireDB2AI: Deep Representation Learning and Prediction Source Technology and Software Development Utilizing Entire Relational Databases Comprehensively', supported by the Institute for Information & Communications Technology Planning & Evaluation (IITP).

Development of a Virtual AI Testbed Capable of Per..

< From left: Professor Jongse Park , M.S candidate Jaehong Cho, M.S candidate Hyunmin Choi, Professor Brandon Reagen ISPASS > Operating Large Language Model (LLM) services like ChatGPT requires a server infrastructure on the scale of tens of thousands of units. However, constructing actual equipment every time a new AI semiconductor or system architecture needs to be verified incurs massive costs and time. A research team at our university has developed a ‘virtual testbed’ that can pre-verify performance and efficiency inside a computer before building an actual large-scale AI server. KAIST announced on May 29th that the research on a Large Language Model (LLM) serving infrastructure simulator (virtual testing software) developed by Professor Jongse Park’s research team in the School of Computing won the Best Paper Award at ‘ISPASS 2026 (IEEE International Symposium on Performance Analysis of Systems and Software),’ a world-renowned conference in the field of computer system performance analysis. ‘LLMServingSim 2.0,’ developed by the research team, is a simulation platform capable of virtually analyzing various hardware and software combinations in complex AI service environments. Researchers and developers can freely experiment with various design options and verify performance without having to directly build expensive, large-scale server infrastructures. < LLMServingSim 2.0 is workload > In particular, this technology is drawing attention because it goes beyond the existing Graphics Processing Unit (GPU)-centric environment to support diverse hardware environments, including Neural Processing Units (NPUs), which are rising as next-generation AI semiconductors, and Processing-In-Memory (PIM, a semiconductor technology that performs operations inside the memory). In other words, it is a technology that allows future-oriented AI semiconductors that have not yet been commercialized to be tested in advance within a virtual datacenter environment. Through this, it is possible to replicate and analyze inside a computer how much the service speed improves, how much power consumption is reduced, and whether it operates stably even in a server environment scaled to tens of thousands of units when a specific semiconductor is applied. In addition, it reproduces complex operations that occur during actual AI service operations—such as data processing, request distribution, and memory utilization—at the system level, enabling performance evaluations that are close to reality. Notably, it can even analyze disaggregated infrastructure environments where multiple server resources are separated and connected for use, showing great potential for utilization in next-generation AI datacenter research. This simulator is expected to be widely utilized not only by researchers but also by LLM service companies and AI semiconductor startups to design and optimize next-generation AI infrastructures. This is because it can rapidly verify new AI semiconductors or service architectures prior to actual construction, thereby significantly reducing the cost and time of AI infrastructure development. < Research Image (AI-generated image) > Professor Jongse Park said, “The competitiveness of AI services is determined not only by the model itself but also by the infrastructure technology that operates it stably and efficiently.” He added, “We hope this simulator will serve as an important foundation for researchers and the industry to develop next-generation AI infrastructures faster and more efficiently.” This research was led by M.S candidate Jaehong Cho and Hyunmin Choi in the School of Computing as co-first authors. Following their Best Paper Award at the 2024 IISWC (IEEE International Symposium on Workload Characterization), the research team won the Best Paper Award again at this ISPASS 2026, proving their research competitiveness in the field of AI infrastructure once more. ※ Paper Title: LLMServingSim 2.0: A Unified Simulator for Heterogeneous and Disaggregated LLM Serving Infrastructure, DOI: 10.1109/ISPASS69572.2026.00012 (Authors: Jaehong Cho, Hyunmin Choi, Guseul Heo, Jongse Park) ※ Open Source Link: https://llmservingsim.ai/ Meanwhile, this research was conducted with support from the Ministry of Science and ICT (MSIT), the Institute for Information & Communications Technology Planning & Evaluation (IITP, No. RS-2024-00396013), the Electronics and Telecommunications Research Institute (ETRI, No. RS-2025-02305453), and SK hynix.

KAIST Develops AI Technology That Automatically Ge..

<(From Left) Hyun-Bin Oh, Takida Yuhta, Uesaka Toshimitsu, Tae-Hyun Oh, Mitsufuji Yuki> When people watch a scene in the film Jurassic Park where a giant dinosaur walks toward them, they naturally imagine a heavy, rumbling sound, as if the ground were shaking. This is because humans predict sound by considering not only the shape of an object, but also physical properties such as its size, weight, and speed of movement. However, existing video-to-audio generation AI mainly generates sound based on the category of objects or scene information in the video, and has not sufficiently reflected physical properties that vary depending on weight or speed. KAIST (President Kwang Hyung Lee) announced on the 26th of May that a collaborative research team involving Professor Tae-Hyun Oh of the School of Computing, KAIST, together with joint researchers from POSTECH (President Sung Keun Kim) and Sony AI, has developed “PAVAS (Physics-Aware Video-to-Audio Synthesis),” an artificial intelligence (AI) technology that understands the physical situation in a video and generates more realistic sound. <Concept Diagram of PAVAS (Physics-Aware Video-to-Audio Synthesis) Technology> The key feature of this technology is that it is designed so that AI can infer invisible physical information such as the mass and velocity of objects in a video on its own. Ordinary videos do not provide exact numerical values for an object’s weight or speed, but the research team enabled AI to estimate them by analyzing the surrounding environment and movement context, and to reflect the results in the sound generation process. In other words, the AI was designed to go beyond simply recognizing “what is visible” and to understand the physical cause of “why this sound should occur.” As a result of technical validation, the research team’s AI generated sounds very similar to real-world environments in scenes involving physical interactions such as collisions or impacts between objects. In particular, it produced more realistic audio in which loudness and tone naturally changed when the mass and velocity of objects varied. Recently, generative AI technologies that simultaneously generate video and audio have been advancing rapidly. Representative examples include Google’s “Veo 3” and ByteDance’s “Seedance 2.0.” However, in actual film, advertising, and game production sites, there is far greater demand for post-production work that adds sound effects suited to existing video scenes or supplements audio than for generating entirely new videos. While existing commercial AI models have focused on generating video and audio together, PAVAS is differentiated by its ability to analyze the movement and collision characteristics of objects in a video and generate realistic sound effects that precisely match the scene. <Comparison of Spectrograms Generated by Conventional Video-to-Audio Models and PAVAS> The research team explained that this technology presents new possibilities in the field of “Physical AI,” or physically consistent generative AI. Physically consistent generative AI refers to AI that goes beyond simply producing plausible results and understands the laws of physics and causal relationships in the real world. In the future, this technology is expected to provide more immersive user experiences in a wide range of fields, including the automation of content sound production, augmented reality (AR) and virtual reality (VR) content, the metaverse, and robotics simulation. Professor Tae-Hyun Oh stated, “While existing generative AI has developed by increasing the scale of data and models, this research is meaningful in that it was designed so that AI directly understands physical quantities and causal relationships,” adding, “In the future, it can be expanded into a core foundational technology for next-generation multimodal AI that simultaneously understands and processes diverse types of information, including text, video, and speech.” This study was led by POSTECH integrated M.S.-Ph.D. student Hyun-Bin Oh as the first author, with KAIST Professor Tae-Hyun Oh and Sony AI researchers Yuhta Takida, Toshimitsu Uesaka, and Yuki Mitsufuji participating as co-authors. This research was selected as an Oral presentation paper at CVPR 2026 (Computer Vision and Pattern Recognition 2026), the world’s most prestigious academic conference in the field of computer vision (image-based artificial intelligence technology), where only the top 0.88％ of all papers are selected for oral presentation, recognizing the excellence of the work. The presentation is scheduled to take place on June 6. ※ Paper title: “PAVAS: Physics-Aware Video-to-Audio Synthesis,” DOI: https://arxiv.org/abs/2512.08282 This research was supported by the Mid-Career Research Program under the Basic Research Program of the Ministry of Science and ICT, the Pioneer Research Program for Future Converging Technology of the Ministry of Science, ICT and Future Planning, the AGI Program of the Ministry of Science and ICT, and the KAIST InnoCORE Program.

Talking to AI Before Seeing a Doctor… KAIST Develo..

<(Front row, from right)Professor Uichin Lee, Professor Eunjoo Kim, Professor Tak Yeon Lee, (Back row, from left) M.S candidate Gyeongmin Na, Ph.D candidate Yugyeong Jung, Researcher Hyangkyeong Oh, M.S candidate Jae Young Choi, Ph.D candidate Hyun Seung Moon> People often say that seeking psychiatric care can feel intimidating. Patients may feel burdened when they first open up about their emotional distress, while medical staff must accurately understand a patient’s extensive history and symptoms within limited consultation time. Korean researchers have developed artificial intelligence (AI) technology that supports the initial psychiatric interview process, the first step in psychiatric care. KAIST (President Kwang Hyung Lee) announced on the 24th of May that a joint research team led by Professor Uichin Lee of the School of Computing and Professor Tak Yeon Lee of the Department of Industrial Design, together with Professor Eunjoo Kim’s team from the Department of Psychiatry at Gangnam Severance Hospital (President Yong-Wook Kim), has developed a large language model (LLM)-based technology to support initial psychiatric interviews. This study was conducted in a way that allows patients to first talk with AI before meeting a doctor, helping them organize their symptoms and condition in advance. <AI Interviewer System Overview Diagram> <Ask-Evaluate-Check-Plan Conversation Flow> The research team designed the system so that AI can adjust the flow of conversation according to patient responses. The AI analyzes patients’ answers in real time by comparing them with specialized medical knowledge in psychiatry and generates the key questions that should be asked next. In particular, this system goes beyond simple question-and-answer interaction by applying real counseling techniques such as expressions of empathy, restating the patient’s words in an organized way, and clarifying ambiguous content. This is intended to help patients talk about their condition more comfortably. As a result of experiments conducted with 1,440 virtual patients to verify performance, the team confirmed that in most cases, the system effectively obtained key clinical information needed for treatment within just 30 minutes. Based on the collected conversation, the AI generates a clinical dashboard that shows symptoms and potential conditions at a glance and provides it to medical staff. Through this, doctors can understand the patient’s condition more systematically before the patient enters the consultation room, allowing them to focus more on in-depth counseling with the patient during the actual consultation. The core of this research is that AI is defined not as a replacement for doctors, but as a “coachable apprentice ” It is a collaborative model in which AI handles repetitive and structured information collection, while doctors make the final diagnosis and prescription based on that information. The research team made clear that AI still has limitations in understanding subtle emotional changes or handling sensitive topics, and emphasized that final judgment must always be carried out by trained medical professionals. Professor Uichin Lee stated, “If AI reduces the burden of the initial consultation stage, medical staff can focus more on deeper counseling with patients,” adding, “This shows the possibility of developing a new model of care in which humans and AI collaborate in medical settings.” This study, with doctoral student Yugyeong Jung as the first author, was presented on April 13 at ACM CHI 2026 (ACM Conference on Human Factors in Computing Systems), the most prestigious conference in the field of human-computer interaction. ※ Paper title: “Toward Flexible Psychiatric History-Taking and Visualization: Exploring Clinician Perspectives with Large Language Models,” DOI: https://dl.acm.org/doi/10.1145/3772318.3790970 ※ Author information: Yugyeong Jung (KAIST, first author), Thu Hoang Anh Vo (KAIST, second author), Hyun Seung Moon (KAIST, third author), Jae Young Choi (KAIST, fourth author), Hyangkyeong Oh (Gangnam Severance Hospital, fifth author), Ujin Lee (Gangnam Severance Hospital, sixth author), Eunjoo Kim (Gangnam Severance Hospital, seventh author), Tak Yeon Lee (KAIST, corresponding author), Uichin Lee (KAIST, corresponding author) This research was supported by the Digital Columbus Project of the Institute of Information & Communications Technology Planning & Evaluation (project title: Development of Digital Innovation Element Technologies for Predicting Complex Diseases in Advance and Expanding Non-Face-to-Face Care).

Overcoming the Limits of Hydrogen Storage and Tran..

<(Top row, from left) Professor Kang Taek Lee, Professor Joongmyeon Bae, Dr. Tae Ho Shin, Dr. Ki-Min Roh, (Bottom row, from left) Dr. Dongyeon Kim, Researcher Dong Jae Park, Dr. Incheol Jeong> As ammonia gains attention as a next-generation energy source capable of overcoming the limits of hydrogen storage and transport, KAIST and a joint research team have developed fuel cell technology that directly uses ammonia as fuel while achieving world-class performance and stability. This achievement is regarded as a core technology that will accelerate the commercialization of the next-generation hydrogen economy and carbon-free power generation. KAIST (President Kwang Hyung Lee) announced on the 20th of May that Professor Kang Taek Lee and Professor Joongmyeon Bae of the Department of Mechanical Engineering, together with a joint research team including Dr. Tae Ho Shin of the Korea Institute of Ceramic Engineering and Technology (KICET, President Jong-Suk Yoon) and Dr. Ki-Min Roh of the Korea Institute of Geoscience and Mineral Resources (KIGAM, President Kwon Yi Kyun), have developed catalyst technology that dramatically improves the performance and durability of ammonia-based protonic ceramic fuel cells (PCFCs, next-generation high-efficiency fuel cells that generate electricity by transporting hydrogen ions). <AI image: A next-generation fuel cell that generates electricity using ammonia (NH₃) as fuel> Ammonia is attracting attention as a next-generation hydrogen carrier (Energy Carrier, a medium that stores and transports hydrogen) because it is easy to store and transport in liquid form. It is also regarded as a representative carbon-free fuel because it consists only of nitrogen (N) and hydrogen (H), producing almost no carbon dioxide (CO₂) during power generation. However, inside fuel cells, ammonia has caused problems by damaging nickel-based materials and slowing reaction rates, leading to performance degradation and shortened lifespan. To solve this problem, the research team designed a new catalyst structure combining a “high-entropy” oxide catalyst (High-Entropy, a design method that enhances material stability and performance by mixing multiple elements) that improves structural stability by mixing multiple elements, with metal nanoparticles (Nano Particle, ultrafine metal particles on the nanometer scale) that form spontaneously on the surface during operation. This catalyst was found not only to resist structural collapse even in an ammonia environment, but also to effectively promote the reaction that decomposes ammonia into hydrogen. Through density functional theory (DFT, Density Functional Theory, a simulation method that calculates reaction mechanisms at the atomic level) analysis, the research team identified that the high-entropy oxide structure lowers the energy barrier required for ammonia decomposition and promotes the formation of metal particles. <AI-generated image of a high-entropy catalyst made by mixing multiple metallic elements> In particular, the metal alloy nanoparticles that formed spontaneously on the catalyst surface showed much higher catalytic activity than single-metal catalysts. A fuel cell applying this catalyst recorded a maximum power density of 2.04 W per unit area (1 cm²) at 700°C. This means that high power can be produced from an area the size of a fingernail, representing world-class performance in the field of ammonia-based protonic ceramic fuel cells that generate electricity by transporting hydrogen ions (protons). In addition, the cell operated stably for more than 255 hours even under harsh conditions of 600°C, significantly improving the problem of performance degradation (a phenomenon in which performance decreases over time) seen in existing catalysts. <Schematic of an ammonia-fueled PCFC incorporating a high-entropy catalyst> <Microstructure and elemental distribution results of the high-entropy catalyst> Professor Kang Taek Lee stated, “Through the synergistic structure of high-entropy oxides and alloy nanoparticles, we improved both the performance and durability of ammonia fuel cells,” adding, “This study will serve as a catalyst for accelerating the commercialization of ammonia-based carbon-free power generation technology and next-generation hydrogen energy systems.” This research, with Dr. Dongyeon Kim of the Department of Mechanical Engineering at KAIST, researcher Dong Jae Park of the Korea Institute of Ceramic Engineering and Technology, and Dr. Incheol Jeong of the Korea Institute of Geoscience and Mineral Resources as co-first authors, was published on April 17 in Nano-Micro Letters (IF: 36.3), an international journal in the fields of energy and materials. ※ Paper title: “Entropy-Modulated Oxide–Metal Catalyst Architectures for Direct Ammonia Protonic Ceramic Fuel Cells,” DOI: https://link.springer.com/article/10.1007/s40820-026-02194-9 This research was supported by the Mid-Career Researcher Program of the Ministry of Science and ICT, the Global Basic Research Laboratory Program, the InnoCORE Program of the Institutes of Science and Technology, and the Basic Research Project of the Korea Institute of Geoscience and Mineral Resources.

KAIST-Hanwha Solutions Establishes ‘Eco-Friendly B..

<(From Left) Hyun Bae Bang, Cheon Woo Moon, Cindy Pricilia Surya Prabowo, Minjung Ki, Sang Yup Lee, Changhee Cho, (Upper Left) Jae Sung Cho, Namjin Jang> KAIST announced on May 19th that the KAIST-Hanwha Solutions Future Technology Research Institute, has secured bio-technology capable of mass-producing eco-friendly raw materials for plastics and textiles using waste resources, offering an alternative to petroleum-derived naphtha. Naphtha, an essential feedstock for the petrochemical industry, has faced sharp price increases and supply instability in recent years, driving demand for sustainable alternatives. The new technology, addresses both resource supply stability and environmental concerns simultaneously. A study led by Distinguished Professor Sang-yup Lee of the Department of Chemical and Biomolecular Engineering was published on May 12th in the journal Nature Chemical Engineering and has been selected as the cover paper for the May issue, a designation reserved for research achievements that represent the corresponding issue. This platform uses ‘glycerol,’ a byproduct discarded during the biodiesel production process, as a raw material. The team engineered high-efficiency microorganisms to convert this waste into 1,3-propanediol (1,3-PDO), a key material for plastics and cosmetics, and optimized the fermentation process for industrial application. The research team succeeded in maintaining high production level even in a 300L pilot process, which serves as a test production stage before application in large-scale plant facilities, moving beyond the laboratory scale. This study also used computer simulations to predict which genes to engineer, which resulted in improved production levels. The team also developed the fermentation system without antibiotic supplementation — a significant advance, as antibiotic use in industrial fermentation raises concerns about antimicrobial resistance and regulatory hurdles for food, cosmetic, and pharmaceutical applications. < (AI Image) Microbial-based process for 1,3-propanediol (1,3-PDO) production > The achievement reflects a 10-year partnership between KAIST and Hanwha Solutions that began in November 2015, with researchers from both sides working together directly on the experiments. Through the KAIST-Hanwha Solutions Future Technology Research Institute, the collaboration has produced 6 patent applications and 13 published papers, standing as a representative model of industry-academic cooperation in South Korea. < Schematic diagram of microbial-based metabolic engineering strategies for 1,3-PDO production > ※ Paper Title: High-titer, antibiotic-free, pilot-scale production of 1,3-propanediol by engineered Corynebacterium, DOI: 10.1038/s44286-026-00389-w ※ Authors: Jae Sung Cho (KAIST, First Author), Cindy Pricilia Surya Prabowo (KAIST, First Author), Taehee Han (KAIST), Cheon Woo Moon (KAIST), Yoo-Sung Ko (KAIST), Changhee Cho (Hanwha Solutions), Je Woong Kim (KAIST), Won Jun Kim (Hanwha Solutions), Hyun Bae Bang (Hanwha Solutions), Jae Eun Lee (KAIST), Minjung Ki (KAIST), Namjin Jang (Hanwha Solutions), Sang Yup Lee (KAIST, Corresponding Author) Jung-dae Kim, Head of the Research Institute at Hanwha Solutions, said, “This research is highly significant in that it confirmed the possibility of replacing existing petrochemical processes using bio-based raw materials. We expect it to be an important foundation for sustainable chemical material production and industrial application in the future.” KAIST Distinguished Professor Sang Yup Lee of the Department of Chemical and Biomolecular Engineering stated, “This research is a case showing that microorganism-based chemical production can be sufficiently expanded to an actual industrial scale beyond the laboratory. It will contribute to producing various chemical materials in a more eco-friendly way in the future.”

KAIST Wins Best Paper Award at Top HCI Conference ..

<(From Left) Professor Chang Hee Lee, Ph.D candidate Yoonji Lee> A new type of digital game has emerged in which plants themselves change characters in the game, while humans observe and emotionally engage with them. KAIST announced on the 15th of May that a research team led by Professor Chang Hee Lee of the Department of Industrial Design won the Best Paper Award at ACM CHI 2026, the most prestigious conference in the field of Human-Computer Interaction (HCI), for research that uses plants not as simple decorations or sensors but as “agents of interaction.” < Plant.play system image > < Plant.play system side view photo > ACM (Association for Computing Machinery) CHI (Conference on Human Factors in Computing Systems) 2026 was held from April 13 to 17 in Barcelona, Spain. CHI is one of the world’s most prestigious international conferences in the field of Human-Computer Interaction (HCI). The Best Paper Award is the highest honor awarded to only about the top 1％ of all submitted papers. In particular, this year’s conference received a total of 6,730 paper submissions, marking the largest scale in its history, and this award is regarded as an achievement demonstrating the global research competitiveness of KAIST researchers. Professor Chang Hee Lee’s team proposed a new form of interaction in which plants directly participate in digital games through the paper “When Plants Play: Rethinking Plant Materiality in Digital Games.” This research is characterized by a design that goes beyond the conventional approach of using plants simply as sensors or decorative elements, allowing changes in the plant’s state to directly affect the progress of the game. The research team reflected the plant’s bioelectrical signals, environmental data, and circadian rhythms (biological changes that repeat according to day and night) in the game, enabling the character in the game to change according to the plant’s state. Rather than directly controlling the game, users participate by observing and interpreting the plant’s changes and responses. As the plant grows, it creates different forms of characters and changes, and these changes reflect the plant’s own growth patterns and pace of transformation. As a result of conducting user research in an actual exhibition environment, the research team confirmed that participants accepted the plant’s slow and unpredictable changes as a form of “play.” In particular, participants also showed a tendency to become emotionally immersed in and empathize with both the plant and the virtual character in the game. This led them to perceive the plant not simply as an object of observation, but as an entity with which they interact. <A digital pet raised by the plant shows various behaviors—such as reading books—according to its daily rhythm, and grows over time> <In a low-humidity environment, the plant provides hamburgers to the digital pet to help it grow> This research received high recognition for moving beyond human-centered digital interaction and proposing new possibilities for interaction with nonhuman entities such as plants. Professor Chang Hee Lee stated, “This research is an attempt to view nonhuman entities such as plants as agents and explore new forms of interaction,” adding, “Our society is expanding into an ‘attachment economy,’ which values emotional bonds and empathy, and in the future, emotional engagement not only with humans but also with diverse nonhuman entities such as AI, robots, animals, and plants will become important.” He continued, “This research is an example that demonstrates these new possibilities for interaction.” This study, with doctoral student Yoonji Lee as the first author and Professor Chang Hee Lee as the corresponding author, can be found in the ACM Digital Library. ※ Paper title: “When Plants Play: Rethinking Plant Materiality in Digital Games” ※ DOI: https://doi.org/10.1145/3772318.3791373 This research was supported by Brain Korea (BK21).

KAIST Develops Real-Time Diagnostic Smart Dressing..

<(From Left) Professor Inkyu Park, Dr. Seokjoo Cho, (Upper Right, From Left) Professor Ji-Hwan Ha, Researcher Junho Jeong , Professor Wei Gao> “Diabetic ulcers,” which occur in patients with diabetes, are dangerous complications that can lead to amputation if the treatment window is missed. A joint research team has developed a “smart dressing patch” that can monitor wound conditions in real time. KAIST (President Kwang Hyung Lee) announced on the 14th of May that a research team led by Distinguished Professor Inkyu Park of the Department of Mechanical Engineering, through joint research with Professor Ji-Hwan Ha of Hanbat National University (President Yongjun Oh), researcher Junho Jeong of the Korea Institute of Machinery & Materials (President Seog-Hyeon Ryu), and Professor Wei Gao of the California Institute of Technology (Caltech; President Thomas F. Rosenbaum) in the United States, has developed a “wireless, battery-free optoelectronic multi-modal sensor patch” for diabetic ulcer management. The patch developed by the research team combines an optoelectronic sensor, which can simultaneously measure multiple types of biological information, with a functional dressing. It can analyze glucose concentration, acidity (pH, an indicator of hydrogen ion concentration), and temperature changes at the wound site in real time, and patients can check their condition themselves using a smartphone. The research team fabricated a functional nanofiber dressing using electrospinning, a method that uses an electric field to create fibers much thinner than a human hair. This dressing changes color in response to increased glucose and changes in acidity that appear in diabetic foot wounds. In other words, if the wound condition worsens, the dressing color changes, allowing danger signals to be easily checked with the naked eye. Through this, abnormal signs that could lead to tissue necrosis can be detected and tracked over long periods in a non-invasive manner, meaning without cutting the skin or drawing blood. The research team combined this with an optoelectronic system to improve diagnostic accuracy. A light-emitting diode (LED, a semiconductor device that converts electricity into light) embedded in the patch and a photodiode, a semiconductor sensor that detects light, measure the color change of the dressing as light reflectance and then convert it into an electrical signal. This provides more accurate and stable data than ordinary camera-based imaging because it is less affected by changes in surrounding lighting. In particular, the patch operates without a separate battery by applying a flexible circuit based on near field communication (NFC), a wireless communication technology that exchanges data over short distances. When a smartphone is placed near the sensor, the patch receives power wirelessly and operates, transmitting the measured data in real time. In other words, patients and medical staff can immediately check and respond to wound conditions using only a smartphone app, without separate complex equipment. < Conceptual Diagram of a Multimodal Colorimetric Dressing and Optoelectronic Sensor for Diagnosing Diabetic Foot and Diabetic Diseases > The technology developed in this study has high clinical value because it provides both intuitive visual signals and quantitative electronic data while imposing no physical burden on patients. It is also expected to contribute to improving the quality of life of patients with diabetes by enabling continuous wound management without repeated blood sampling. Distinguished Professor Inkyu Park stated, “Research that began to reduce the pain of diabetic patients who have to prick their fingers with a needle every day has led to a technology for the preemptive diagnosis of complications,” adding, “This technology will become a core platform technology that can be expanded in the future to blood-free diagnostic technologies not only for diabetes but also for various chronic diseases.” In this study, KAIST Dr. Seokjoo Cho and Professor Ji-Hwan Ha of Hanbat National University participated as co-first authors. The research results were published on March 26, 2026, in the international materials science journal Advanced Functional Materials. The paper was also selected as a Front Cover article of the journal. ※ Paper title: “Wireless, Battery-Free, Optoelectronic, Multi-Modal Sensor Integrated With Colorimetric Dressing for Diabetic Ulcer Management,” DOI: 10.1002/adfm.202532167 < Front Cover Image > This research was supported by the Mid-Career Researcher Program of the National Research Foundation of Korea (NRF) funded by the Ministry of Science and ICT, the Alchemist Project of the Ministry of Trade, Industry and Energy, and the Daejeon RISE Center.

“Even Recognizing Puddles at Night”... KAIST Surpa..

< (From left) Ph.D. student Hanbin Cho, Postdoctoral Researcher Wenxuan Zhu, Professor Joonki Suh, and MS-PhD integrated student Changhwan Kim > A technology that surpasses the limitations of existing sensors, which failed to distinguish between water and asphalt on dark roads, has emerged to enhance the accuracy of autonomous driving and medical diagnostics. Our university's research team has developed a next-generation polarization sensor that can read the "direction" of light and change its own response. KAIST announced on May 12th that a research team led by Professor Joonki Suh from the Department of Chemical and Biomolecular Engineering has developed a "self-reconfigurable" polarization sensor array technology that regulates its operation by finding the optimal state using "polarization" information—the property of light vibrating in a specific direction. With the recent explosive increase in data and the rapid development of artificial intelligence technology, the need for next-generation vision systems that can efficiently process vast amounts of information with low energy is growing. However, existing image sensors only detect the intensity (brightness) of light, limiting their ability to precisely grasp the orientation or surface structure of objects. To overcome these limitations, the research team developed a polarization-based sensor technology capable of recognizing the vibration direction of light. In particular, by utilizing a "heterostructure" that combines two different materials—tellurium (Te) and rhenium disulfide (ReS₂)—they effectively implemented characteristics where the response to light varies depending on the crystal orientation. < Conceptual diagram of self-reconfigurable polarization sensor and in-sensor computing based on dual-anisotropy vdW heterostructures > To precisely stack the two materials so they cross each other, the research team applied "Epitaxial Atomic Layer Deposition," a process that controls crystal structures by stacking materials precisely at the atomic layer level. By ensuring the crystal structures of the two materials interlock accurately, they secured higher reproducibility and stable performance compared to previous methods. In this structure, when light is irradiated, interfacial carrier transfer and trapping (a phenomenon where electrons move or stay at specific locations) occur at the material boundary. As a result, a "bipolar photoresponse"—a light-induced reaction where the current direction flips depending on conditions such as light intensity, wavelength, and direction—appears. A key feature is that the sensor's operating state can be freely adjusted using only light, without external electrical signals. Furthermore, this technology can be applied to "in-sensor computing" structures where the sensor itself processes data, allowing for the efficient processing of multi-dimensional optical information that changes over time without complex calculation processes. In actual experiments, it recorded a high accuracy of over 95％ in recognizing moving objects, proving its potential for applications in various fields such as autonomous driving and medical diagnosis. < Experimental image of a polarization AI sensor platform capable of light-based operational reconfiguration (AI-generated image) > Professor Joonki Suh stated, "This research presents a new foundation for AI vision technology that can secure richer visual information by utilizing polarization information. It is expected to play an important role in implementing low-power, high-efficiency AI systems in the future." Wenxuan Zhu (Postdoctoral Researcher) and Changhwan Kim (Ph.D. student) participated as first authors in this study, with Professor Joonki Suh participating as the corresponding author. The research results were published on April 14 in the international academic journal Nature Sensors. Paper Title: Self-reconfigurable polarization perception in dual-anisotropy heterostructures for high-dimensional in-sensor computing Authors: Wenxuan Zhu, Changhwan Kim, Ruofan Zhang, Mingchun Lu, Namwook Hur, Hanbin Cho, Jihyun Kim, Jiacheng Sun, Joohoon Kang, Junchi Yan, Yuan Cheng & Joonki Suh DOI: https://doi.org/10.1038/s44460-026-00057-9 < Paper portfolio and QR code > Meanwhile, this research was conducted with the support of the PIM AI Semiconductor Core Technology Development (Device) Project and the Individual Basic Research Project of the National Research Foundation of Korea, funded by the Ministry of Science and ICT, and the Industrial Innovation Talent Growth Support Project of the Korea Institute for Advancement of Technology (KIAT).