%52%55%42
-
Physics Informed AI Excels at Large Scale Discovery of New Materials!
<(From left) Ph.D candidates Songho Lee, Donggeun Park, and Hyeonbin Moon, and Professor Seunghwa Ryu from the Department of Mechanical Engineering; (top) Professor Jae Hyuk Lim from Kyung Hee University and Dr. Wabi Demeke from KAIST>
One of the key steps in developing new materials is “property identification,” which has long relied on massive amounts of experimental data and expensive equipment, limiting research efficiency. A KAIST research team has introduced a new technique that combines “physical laws,” which govern deformation and interaction of materials and energy, with artificial intelligence. This approach allows for rapid exploration of new materials even under data-scarce conditions and provides a foundation for accelerating design and verification across multiple engineering fields, including materials, mechanics, energy, and electronics.
KAIST (President Kwang Hyung Lee) announced on the 2nd of October that Professor Seunghwa Ryu’s research group in the Department of Mechanical Engineering, in collaboration with Professor Jae Hyuk Lim’s group at Kyung Hee University (President Jinsang Kim) and Dr. Byungki Ryu at the Korea Electrotechnology Research Institute (President Namkyun Kim), proposed a new method that can accurately determine material properties with only limited data. The method uses Physics-Informed Machine Learning (PIML), which directly incorporates physical laws into the AI learning process.
<Schematic Diagram of a Physics-Based Machine Learning Methodology for Understanding Material Properties>
In the first study, the researchers focused on hyperelastic materials, such as rubber. They presented a Physics-Informed Neural Network (PINN) method that can identify both the deformation behavior and the properties of materials using only a small amount of data obtained from a single experiment. Whereas previous approaches required large, complex datasets, this research demonstrated that material characteristics can be reliably reproduced even when data is scarce, limited, or noisy.
In the second study, the team turned to thermoelectric materials—new materials that convert heat into electricity and electricity into heat. They proposed a PINN-based inverse inference technique that can estimate key indicators, such as thermal conductivity (how well heat is transferred) and the Seebeck coefficient (how efficiently electricity is generated), from just a few measurements.
Going further, the researchers introduced a Physics-Informed Neural Operator (PINO), an AI model that understands the physical laws of nature, and showed that it can generalize to previously unseen materials without requiring retraining.
In fact, after training the system on 20 materials, they tested it on 60 entirely new materials, and in all cases it predicted their properties with high accuracy. This breakthrough points to a future where large-scale, high-speed screening of countless candidate materials becomes possible.
This achievement goes beyond simply reducing the need for experiments. By intricately combining physical laws with AI, the researchers provided the first example of improving experimental efficiency while preserving reliability.
Professor Seunghwa Ryu, who led both studies, stated, “This is the first case of applying AI that understands physical laws to real material research. It enables reliable identification of material properties even when data availability is limited, and it is expected to expand into various engineering fields.”
The first paper, co-first-authored by KAIST Mechanical Engineering PhD candidates Hyeonbin Moon and Donggeun Park, was published on August 13 in Computer Methods in Applied Mechanics and Engineering.
※ Paper title: “Physics-informed neural network-based discovery of hyperelastic constitutive models from extremely scarce data”
※ DOI: https://doi.org/10.1016/j.cma.2025.118258
The second paper, co-first-authored by KAIST Mechanical Engineering PhD candidates Hyeonbin Moon and Songho Lee, and Dr. Wabi Demeke, was published on August 22 in npj Computational Materials.
※ Paper title: “Physics-informed neural operators for generalizable and label-free inference of temperature-dependent thermoelectric properties” ※ DOI: https://doi.org/10.1038/s41524-025-01769-1
Meanwhile, the first study was supported by the Korea Research Foundation and the Ministry of Science and ICT’s INNOCore Program, as well as by a research project from the Ministry of Food and Drug Safety. The second study was carried out with support from the Korea Research Foundation and the Ministry of Science and ICT’s INNOCore Program.
2025.10.10 View 2448 -
In KAIST, Robots Now Untie Rubber Bands and Insert Wires Like Humans
The technology that allows robots to handle deformable objects such as wires, clothing, and rubber bands has long been regarded as a key task in the automation of manufacturing and service industries. However, since such deformable objects do not have a fixed shape and their movements are difficult to predict, robots have faced great difficulties in accurately recognizing and manipulating them. KAIST researchers have developed a robot technology that can precisely grasp the state of deformable objects and handle them skillfully, even with incomplete visual information. This achievement is expected to contribute to intelligent automation in various industrial and service fields, including cable and wire assembly, manufacturing that handles soft components, and clothing organization and packaging.
KAIST (President Kwang Hyung Lee) announced on the 21st of August that the research team led by Professor Daehyung Park of the School of Computing developed an artificial intelligence technology called “INR-DOM (Implicit Neural-Representation for Deformable Object Manipulation),” which enables robots to skillfully handle objects whose shape continuously changes like elastic bands and which are visually difficult to distinguish.
Professor Park’s research team developed a technology that allows robots to completely reconstruct the overall shape of a deformable object from partially observed three-dimensional information and to learn manipulation strategies based on it. Additionally, the team introduced a new two-stage learning framework that combines reinforcement learning and contrastive learning so that robots can efficiently learn specific tasks. The trained controller achieved significantly higher task success rates compared to existing technologies in a simulation environment, and in real robot experiments, it demonstrated a high level of manipulation capability, such as untying complicatedly entangled rubber bands, thereby greatly expanding the applicability of robots in handling deformable objects.
Deformable Object Manipulation (DOM) is one of the long-standing challenges in robotics. This is because deformable objects have infinite degrees of freedom, making their movements difficult to predict, and the phenomenon of self-occlusion, in which the object hides parts of itself, makes it difficult for robots to grasp their overall state.
To solve these problems, representation methods of deformable object states and control technologies based on reinforcement learning have been widely studied. However, existing representation methods could not accurately represent continuously deforming surfaces or complex three-dimensional structures of deformable objects, and since state representation and reinforcement learning were separated, there was a limitation in constructing a suitable state representation space needed for object manipulation.
To overcome these limitations, the research team utilized “Implicit Neural Representation.” This technology receives partial three-dimensional information (point cloud*) observed by the robot and reconstructs the overall shape of the object, including unseen parts, as a continuous surface (signed distance function, SDF). This enables robots to imagine and understand the overall shape of the object just like humans.
*Point cloud 3D information: a method of representing the three-dimensional shape of an object as a “set of points” on its surface.
Furthermore, the research team introduced a two-stage learning framework. In the first stage of pre-training, a model is trained to reconstruct the complete shape from incomplete point cloud data, securing a state representation module that is robust to occlusion and capable of well representing the surfaces of stretching objects. In the second stage of fine-tuning, reinforcement learning and contrastive learning are used together to optimize the control policy and state representation module so that the robot can clearly distinguish subtle differences between the current state and the goal state and efficiently find the optimal action required for task execution.
When the INR-DOM technology developed by the research team was mounted on a robot and tested, it showed overwhelmingly higher success rates than the best existing technologies in three complex tasks in a simulation environment: inserting a rubber ring into a groove (sealing), installing an O-ring onto a part (installation), and untying tangled rubber bands (disentanglement). In particular, in the most challenging task, disentanglement, the success rate reached 75%, which was about 49% higher than the best existing technology (ACID, 26%).
The research team also verified that INR-DOM technology is applicable in real environments by combining sample-efficient robotic reinforcement learning with INR-DOM and performing reinforcement learning in a real-world environment.
As a result, in actual environments, the robot performed insertion, installation, and disentanglement tasks with a success rate of over 90%, and in particular, in the visually difficult bidirectional disentanglement task, it achieved a 25% higher success rate compared to existing image-based reinforcement learning methods, proving that robust manipulation is possible despite visual ambiguity.
Minseok Song, a master’s student and first author of this research, stated that “this research has shown the possibility that robots can understand the overall shape of deformable objects even with incomplete information and perform complex manipulation based on that understanding.” He added, “It will greatly contribute to the advancement of robot technology that performs sophisticated tasks in cooperation with humans or in place of humans in various fields such as manufacturing, logistics, and medicine.”
This study, with KAIST School of Computing master’s student Minseok Song as first author, was presented at the top international robotics conference, Robotics: Science and Systems (RSS) 2025, held June 21–25 at USC in Los Angeles.
※ Paper title: “Implicit Neural-Representation Learning for Elastic Deformable-Object Manipulations”
※ DOI: https://www.roboticsproceedings.org/ (to be released), currently https://arxiv.org/abs/2505.00500
This research was supported by the Ministry of Science and ICT through the Institute of Information & Communications Technology Planning & Evaluation (IITP)’s projects “Core Software Technology Development for Complex-Intelligence Autonomous Agents” (RS-2024-00336738; Development of Mission Execution Procedure Generation Technology for Autonomous Agents’ Complex Task Autonomy), “Core Technology Development for Human-Centered Artificial Intelligence” (RS-2022-II220311; Goal-Oriented Reinforcement Learning Technology for Multi-Contact Robot Manipulation of Everyday Objects), “Core Computing Technology” (RS-2024-00509279; Global AI Frontier Lab), as well as support from Samsung Electronics. More details can be found at https://inr-dom.github.io.
2025.08.21 View 3738 -
New and Highly Efficient Recycling Technology to Turn Used Tires into Raw Materials for Rubber and Nylon
< (From left) Kyungmin Choi (MS-Ph.D. integrated course, Department of Chemistry), Dr. Beomsoon Park, Professor Soon Hyeok Hong, Dr. Kyoungil Cho >
Approximately 1.5 billions of tires are discarded globally every year, and this is identified as one of the major causes of serious environmental pollution. The research team at the Department of Chemistry at KAIST has achieved a breakthrough by selectively converting waste tires into high-purity cyclic alkenes, valuable chemical building blocks used in the production of rubber and nylon fibers. This advance marks a new milestone in chemical recycling technology for waste tires.
The team, led by Professor Soon Hyeok Hong, has developed a dual-catalyst-based reaction system that overcomes the long-standing challenges associated with recycling vulcanized rubber materials.
Tires are composed of complex blends of synthetic and natural rubber, and their physical strength and durability are reinforced with additives such as silica, carbon black, and antioxidants. In particular, cross-linking between rubber chains is formed through the vulcanization process, giving them a structure resistant to heat and pressure, which is one of the main reasons why chemical recycling of waste tires is difficult.
Until now, waste tire recycling has mainly relied on pyrolysis or mechanical recycling methods. The pyrolysis method is a technology that decomposes polymer chains at high temperatures of 350-800°C to convert them into fuel oil, but it clearly has limitations such as high energy consumption, low selectivity, and the production of low-quality hydrocarbon mixtures.
To solve these problems, the research team developed a method to convert waste rubber into useful chemicals using dual catalysis. The first catalyst helps to break down rubber molecules by changing their bonding structure, and the second catalyst creates cyclic compounds through a ring-closing reaction.
This process shows high selectivity of up to 92% and a yield of 82%. The produced cyclopentene can be recycled into rubber, and cyclohexene can be used as a raw material for nylon fibers, making them industrially very valuable.
The research team successfully applied the developed system to discarded waste tires, achieving selective conversion into high-purity cyclic alkenes. Unlike the existing pyrolysis method, this is evaluated as a new turning point in the field of waste tire recycling as it can produce high-value chemicals through low-temperature precision catalytic reactions.
In addition, this catalytic platform is compatible with a wide range of synthetic and waste rubbers, positioning it as a promising foundation for scalable, circular solutions in the polymer and materials industries.
< Figure 1. Development of a Catalytic Method for Chemical Recycling of Waste Rubber >
Professor Hong stated, "This research offers an innovative solution for the chemical recycling of waste tires. We aim to develop next-generation high-efficiency catalysts and lay the groundwork for commercialization to enhance economic feasibility. Ultimately, our goal is to contribute to solving the broader waste plastic problem through fundamental chemistry."
This research, in which Beomsoon Park, Kyoungil Cho, and Kyungmin Choi participated, was supported by the National Research Foundation of Korea and was published online in the internationally renowned academic journal ‘Chem’ on June 18th.
※Paper Title: Catalytic and Selective Chemical Recycling of Post-Consumer Rubbers into Cycloalkenes
※DOI: 10.1016/j.chempr.2025.102625
2025.06.26 View 7168 -
A Way for Smartwatches to Detect Depression Risks Devised by KAIST and U of Michigan Researchers
- A international joint research team of KAIST and the University of Michigan developed a digital biomarker for predicting symptoms of depression based on data collected by smartwatches
- It has the potential to be used as a medical technology to replace the economically burdensome fMRI measurement test
- It is expected to expand the scope of digital health data analysis
The CORONA virus pandemic also brought about a pandemic of mental illness. Approximately one billion people worldwide suffer from various psychiatric conditions. Korea is one of more serious cases, with approximately 1.8 million patients exhibiting depression and anxiety disorders, and the total number of patients with clinical mental diseases has increased by 37% in five years to approximately 4.65 million. A joint research team from Korea and the US has developed a technology that uses biometric data collected through wearable devices to predict tomorrow's mood and, further, to predict the possibility of developing symptoms of depression.
< Figure 1. Schematic diagram of the research results. Based on the biometric data collected by a smartwatch, a mathematical algorithm that solves the inverse problem to estimate the brain's circadian phase and sleep stages has been developed. This algorithm can estimate the degrees of circadian disruption, and these estimates can be used as the digital biomarkers to predict depression risks. >
KAIST (President Kwang Hyung Lee) announced on the 15th of January that the research team under Professor Dae Wook Kim from the Department of Brain and Cognitive Sciences and the team under Professor Daniel B. Forger from the Department of Mathematics at the University of Michigan in the United States have developed a technology to predict symptoms of depression such as sleep disorders, depression, loss of appetite, overeating, and decreased concentration in shift workers from the activity and heart rate data collected from smartwatches.
According to WHO, a promising new treatment direction for mental illness focuses on the sleep and circadian timekeeping system located in the hypothalamus of the brain, which directly affect impulsivity, emotional responses, decision-making, and overall mood.
However, in order to measure endogenous circadian rhythms and sleep states, blood or saliva must be drawn every 30 minutes throughout the night to measure changes in the concentration of the melatonin hormone in our bodies and polysomnography (PSG) must be performed. As such treatments requires hospitalization and most psychiatric patients only visit for outpatient treatment, there has been no significant progress in developing treatment methods that take these two factors into account. In addition, the cost of the PSG test, which is approximately $1000, leaves mental health treatment considering sleep and circadian rhythms out of reach for the socially disadvantaged.
The solution to overcome these problems is to employ wearable devices for the easier collection of biometric data such as heart rate, body temperature, and activity level in real time without spatial constraints. However, current wearable devices have the limitation of providing only indirect information on biomarkers required by medical staff, such as the phase of the circadian clock.
The joint research team developed a filtering technology that accurately estimates the phase of the circadian clock, which changes daily, such as heart rate and activity time series data collected from a smartwatch. This is an implementation of a digital twin that precisely describes the circadian rhythm in the brain, and it can be used to estimate circadian rhythm disruption.
< Figure 2. The suprachiasmatic nucleus located in the hypothalamus of the brain is the central biological clock that regulates the 24-hour physiological rhythm and plays a key role in maintaining the body’s circadian rhythm. If the phase of this biological clock is disrupted, it affects various parts of the brain, which can cause psychiatric conditions such as depression. >
The possibility of using the digital twin of this circadian clock to predict the symptoms of depression was verified through collaboration with the research team of Professor Srijan Sen of the Michigan Neuroscience Institute and Professor Amy Bohnert of the Department of Psychiatry of the University of Michigan.
The collaborative research team conducted a large-scale prospective cohort study involving approximately 800 shift workers and showed that the circadian rhythm disruption digital biomarker estimated through the technology can predict tomorrow's mood as well as six symptoms, including sleep problems, appetite changes, decreased concentration, and suicidal thoughts, which are representative symptoms of depression.
< Figure 3. The circadian rhythm of hormones such as melatonin regulates various physiological functions and behaviors such as heart rate and activity level. These physiological and behavioral signals can be measured in daily life through wearable devices. In order to estimate the body’s circadian rhythm inversely based on the measured biometric signals, a mathematical algorithm is needed. This algorithm plays a key role in accurately identifying the characteristics of circadian rhythms by extracting hidden physiological patterns from biosignals. >
Professor Dae Wook Kim said, "It is very meaningful to be able to conduct research that provides a clue for ways to apply wearable biometric data using mathematics that have not previously been utilized for actual disease management." He added, "We expect that this research will be able to present continuous and non-invasive mental health monitoring technology. This is expected to present a new paradigm for mental health care. By resolving some of the major problems socially disadvantaged people may face in current treatment practices, they may be able to take more active steps when experiencing symptoms of depression, such as seeking counsel before things get out of hand."
< Figure 4. A mathematical algorithm was devised to circumvent the problems of estimating the phase of the brain's biological clock and sleep stages inversely from the biodata collected by a smartwatch. This algorithm can estimate the degree of daily circadian rhythm disruption, and this estimate can be used as a digital biomarker to predict depression symptoms. >
The results of this study, in which Professor Dae Wook Kim of the Department of Brain and Cognitive Sciences at KAIST participated as the joint first author and corresponding author, were published in the online version of the international academic journal npj Digital Medicine on December 5, 2024. (Paper title: The real-world association between digital markers of circadian disruption and mental health risks) DOI: 10.1038/s41746-024-01348-6
This study was conducted with the support of the KAIST's Research Support Program for New Faculty Members, the US National Science Foundation, the US National Institutes of Health, and the US Army Research Institute MURI Program.
2025.01.20 View 16397 -
Scientists Observe the Elusive Kondo Screening Cloud
Scientists ended a 50-year quest by directly observing a quantum phenomenon
An international research group of Professor Heung-Sun Sim has ended a 50-year quest by directly observing a quantum phenomenon known as a Kondo screening cloud. This research, published in Nature on March 11, opens a novel way to engineer spin screening and entanglement. According to the research, the cloud can mediate interactions between distant spins confined in quantum dots, which is a necessary protocol for semiconductor spin-based quantum information processing. This spin-spin interaction mediated by the Kondo cloud is unique since both its strength and sign (two spins favor either parallel or anti-parallel configuration) are electrically tunable, while conventional schemes cannot reverse the sign.
This phenomenon, which is important for many physical phenomena such as dilute magnetic impurities and spin glasses, is essentially a cloud that masks magnetic impurities in a material. It was known to exist but its spatial extension had never been observed, creating controversy over whether such an extension actually existed.
Magnetism arises from a property of electrons known as spin, meaning that they have angular momentum aligned in one of either two directions, conventionally known as up and down. However, due to a phenomenon known as the Kondo effect, the spins of conduction electrons—the electrons that flow freely in a material—become entangled with a localized magnetic impurity, and effectively screen it. The strength of this spin coupling, calibrated as a temperature, is known as the Kondo temperature.
The size of the cloud is another important parameter for a material containing multiple magnetic impurities because the spins in the cloud couple with one another and mediate the coupling between magnetic impurities when the clouds overlap. This happens in various materials such as Kondo lattices, spin glasses, and high temperature superconductors.
Although the Kondo effect for a single magnetic impurity is now a text-book subject in many-body physics, detection of its key object, the Kondo cloud and its length, has remained elusive despite many attempts during the past five decades. Experiments using nuclear magnetic resonance or scanning tunneling microscopy, two common methods for understanding the structure of matter, have either shown no signature of the cloud, or demonstrated a signature only at a very short distance, less than 1 nanometer, so much shorter than the predicted cloud size, which was in the micron range.
In the present study, the authors observed a Kondo screening cloud formed by an impurity defined as a localized electron spin in a quantum dot—a type of “artificial atom”—coupled to quasi-one-dimensional conduction electrons, and then used an interferometer to measure changes in the Kondo temperature, allowing them to investigate the presence of a cloud at the interferometer end.
Essentially, they slightly perturbed the conduction electrons at a location away from the quantum dot using an electrostatic gate. The wave of conducting electrons scattered by this perturbation returned back to the quantum dot and interfered with itself. This is similar to how a wave on a water surface being scattered by a wall forms a stripe pattern. The Kondo cloud is a quantum mechanical object which acts to preserve the wave nature of electrons inside the cloud.
Even though there is no direct electrostatic influence of the perturbation on the quantum dot, this interference modifies the Kondo signature measured by electron conductance through the quantum dot if the perturbation is present inside the cloud. In the study, the researchers found that the length as well as the shape of the cloud is universally scaled by the inverse of the Kondo temperature, and that the cloud’s size and shape were in good agreement with theoretical calculations.
Professor Sim at the Department of Physics proposed the method for detecting the Kondo cloud in the co-research with the RIKEN Center for Emergent Matter Science, the City University of Hong Kong, the University of Tokyo, and Ruhr University Bochum in Germany.
Professor Sim said, “The observed spin cloud is a micrometer-size object that has quantum mechanical wave nature and entanglement. This is why the spin cloud has not been observed despite a long search. It is remarkable in a fundamental and technical point of view that such a large quantum object can now be created, controlled, and detected.
Dr. Michihisa Yamamoto of the RIKEN Center for Emergent Matter Science also said, “It is very satisfying to have been able to obtain real space image of the Kondo cloud, as it is a real breakthrough for understanding various systems containing multiple magnetic impurities. The size of the Kondo cloud in semiconductors was found to be much larger than the typical size of semiconductor devices.”
Publication:
Borzenets et al. (2020) Observation of the Kondo screening cloud. Nature, 579. pp.210-213. Available online at https://doi.org/10.1038/s41586-020-2058-6
Profile:
Heung-Sun Sim, PhD
Professor
hssim@kaist.ac.kr
https://qet.kaist.ac.kr/
Quantum Electron Correlation & Transport Theory Group (QECT Lab)
https://qc.kaist.ac.kr/index.php/group1/
Center for Quantum Coherence In COndensed Matter
Department of Physics
https://www.kaist.ac.kr
Korea Advanced Institute of Science and Technology (KAIST)
Daejeon, Republic of Korea
2020.03.13 View 23903