Seoul National University · Engineering
Professor Dohyeong Kim's research lab specializes in safe and efficient reinforcement learning (RL) for robotic systems, with a focus on addressing complex, sequential tasks through intuitive reward shaping and constraint satisfaction. The lab develops advanced RL algorithms that integrate entropy maximization—particularly Tsallis entropy—for improved exploration while maintaining safety through predictive safeguard mechanisms. Their work also extends to environmental systems, particularly coastal morphodynamics, where they study sediment transport and morphological changes in tidal and fluvial environments using high-resolution remote sensing and field data. The lab bridges AI-driven decision making with real-world physical constraints in both robotics and natural systems.
Figures are computed from collected data and may differ slightly.
This study integrates global data sets for aerosols, cloud physical properties, and shortwave radiation fluxes with a Monte Carlo Aerosol‐Cloud‐Radiation (MACR) model to estimate both the surface and the top‐of‐atmosphere (TOA) solar radiation budget as well as atmospheric column solar absorption. The study also quantifies the radiative forcing of aerosols and that of clouds. The observational input to MACR includes data from the Multiangle Imaging Spectroradiometer (MISR) for aerosol optical de
This study analyzes the effects of decoupling of telework on job satisfaction using the 2013 Federal Employee Viewpoint Survey. The research divides telework programs for public employees by two criteria: (a) whether or not federal agencies have officially adopted the program, and (b) whether or not public employees actually participate in the program. We find that both organizational adoption and employee participation in telework have a positive relationship with job satisfaction, and these re
Geo-Kompsat-2A (Geostationary-Korean Multi-Purpose SATtellite-2A, GK2A), a new generation of Korean geostationary meteorological satellite, carry state-of-the-art optical sensors with significantly higher radiometric, spectral, and spatial resolution than the Communication, Ocean, and Meteorological Satellite (COMS) previously available in the geostationary orbit. The new Advanced Meteorological Imager (AMI) on GK2A has 16 observation channels, and its spatial resolution is 0.5 or 1 km for visib
Long-lasting insecticidal nets (LLINs) have been widely used as an effective alternative to conventional insecticide-treated nets (ITNs) for over a decade. Due to the growing number of field trials and interventions reporting the effectiveness of LLINs in controlling malaria, there is a need to systematically review the literature on LLINs and ITNs to examine the relative effectiveness and characteristics of both insecticide nettings. A systematic review of over 2000 scholarly articles published
Despite ample research on the association between indoor air pollution and allergic disease prevalence, public health and environmental policies still lack predictive evidence for developing a preventive guideline for patients or vulnerable populations mostly due to limitation of real-time big data and model predictability. Recent popularity of IoT and machine learning techniques could provide enabling technologies for collecting real-time big data and analyzing them for more accurate prediction
Red quasars are very red in the optical through near-infrared (NIR) wavelengths, which is possibly due to dust extinction in their host galaxies as expected in a scenario in which red quasars are an intermediate population between merger-driven star-forming galaxies and unobscured type 1 quasars. However, alternative mechanisms also exist to explain their red colors: (i) an intrinsically red continuum; (ii) an unusual high covering factor of the hot dust component, that is, CF HD = L HD ∕ L bol
In spite of the increasing efforts to gather spatial data in developing countries, the use of maps is mostly for visualization of health indicators rather than informed decision-making. Various spatial tools can aid policymakers to allocate resources effectively, predict patterns in communicable or infectious diseases, and provide insights into geographical factors which are associated with utilization or adequacy of health services. In Bangladesh, the launch of District Health Information Syste
Indoor air quality becomes more critical as people stay indoors longer, particularly children and the elderly who are vulnerable to air pollution. Natural ventilation has been recognized as the most economical and effective means of improving indoor air quality, but its benefit is questionable when the external air quality is unacceptable. Such risk-risk tradeoffs would require evidence-based guidelines for households and policymakers, but there is a lack of research that examines spatiotemporal
A GIS-based analysis demonstrated several hotspots as potential "food deserts" tailored to the Korean context based on the observed spatial patterns of undesirable food consumption. These findings could be used as a guide to prioritize areas for targeted intervention programs to facilitate healthy food consumption behaviors and thus improve nutrition and food-related health outcomes.
Although previous ecological studies investigating the association between air pollution and allergic diseases accounted for temporal or seasonal relationships, few studies address spatial non-stationarity or autocorrelation explicitly. Our objective was to examine bivariate correlation between outdoor air pollutants and the prevalence of allergic diseases, highlighting the limitation of a non-spatial correlation measure, and suggesting an alternative to address spatial autocorrelation. The 5-ye
As safety is of paramount importance in robotics, reinforcement learning that reflects safety, called <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">safe RL</i> , has been studied extensively. In safe RL, we aim to find a policy which maximizes the desired return while satisfying the defined safety constraints. There are various types of constraints, among which constraints on conditional value at risk (CVaR) effectively lower the probability of
Open papers in the app to read, cite, and organize with AI.
