FAKE Lab.

Cloth–liquid interaction involves coupled phenomena such as porous flow, absorption, emission, and diffusion, which are difficult to model in a stable and efficient manner in particle-based simulations. In this paper, we present a GPU-based cloth–liquid coupling framework that integrates these processes within a unified Smoothed Particle Hydrodynamics (SPH) pipeline. To improve directional robustness in porous flow near boundaries, we use a virtual-pressure formulation inspired by Darcy’s law, while absorption and emission are handled through saturation-based mass exchange and diffusion is applied as a sequential redistribution stage. For efficient neighbor search, the framework employs a Bitonic sort–based GPU hashing structure. We evaluate the proposed method using qualitative wet-cloth scenarios and quantitative analyses including frame-time measurement, sorting-cost comparison, scalability with increasing particle counts, mass conservation error, and module-wise timing breakdown. The results show that the proposed hashing scheme reduces sorting overhead compared with the baseline GPU implementation, while the full coupling pipeline maintains stable wetting behavior and bounded mass variation across the tested scenes. These results indicate that the proposed framework provides an efficient and numerically stable approach for GPU-based porous cloth–liquid simulation.

We present an end-to-end pipeline that delivers gyrokinetic scalar fields into Unreal Engine and NVIDIA Omniverse, with three key advances. The first is the render time, native coordinate sampling of curvilinear structured grid fields, as exemplified by gKPSP on (r, θ, φ), in both real-time engines, removing the intermediate Cartesian resampling pass. The second is the joint use of OpenVDB and OpenUSD as an engine-neutral intermediate representation between the solver output and the two engine front ends. The third is an interactive in- engine inspection through threshold window filtering, clip box dissolution, transfer function tuning, and arbitrary cross-section planes, with the native coordinate scene sustaining interactive rendering on a single graphics processing unit workstation.

The enabling mechanism is an inverse engine XYZ–to-toroidal map evaluated at every ray sample, returning each sample to the simulation frame before the structured grid lookup. The same map was implemented independently in the material, scientific renderer, and compute kernel layers of the two engines. A quantitative comparison against the Cartesian bounding box baselines showed that the native coordinate representation preserves the structured grid resolution and fine-scale turbulence content, which Cartesian resampling discards, at the cost of a bounded per-sample coordinate conversion overhead. The rendering of fields defined on unstructured high-order meshes is left to future work.

In this study, we develop a numerical method to represent turbulent flow in various 2D vector fields using the Monte Carlo method-based MLS (Moving Least Squares) from a density field and express it as a learning representation through a neural network. Conventional MLS performs high-order interpolation solely based on vector-based constraints, making it difficult to effectively reflect the characteristics of a density field, which limits its applicability in various fields. Additionally, equations of higher degree require significant computational effort, making them costly in terms of computation time and resources when applied to simulation tasks that require per-frame calculations. To address these issues, this study integrates the Monte Carlo method-based weighting into MLS to efficiently consider the characteristics of the density field in the input data and design an algorithm that represents it as various forms of vector fields. Furthermore, we extend the solver to express this approach as a learning representation through a neural network. To validate the applicability of our method, we conducted experiments extracting turbulent vector fields from various density fields. Since conventional MLS does not guarantee temporal continuity, directly applying it to simulations results in noise. To resolve this, our method generates turbulent flow by analyzing the angular variation between the generated velocity and the underlying fluid velocity, ensuring stable advection of the density. As a result, our method efficiently and accurately extracts turbulent flow from a density field and integrates it into an underlying fluid solver, enabling the practical use of high-order interpolation in physics-based simulations. Experimental results across various scenarios demonstrate that our method improves both computation time and quality compared to previous methods, yielding enhanced turbulent flow fields.

본 연구는 Microsoft HoloLens 2를 활용하여 현실 물체를 기반으로 사용자 맞춤형 레이싱 트랙을 자동 생성하고, 두 명의 사용자가 동일 공간에서 증강현실(AR) 멀티플레이 레이싱을 수행할 수 있는 ARSlope 시스템을 제안한다. 시스템은 HoloLens 2의 공간 인식을 통해 수평 평면과 그 위의 물체를 인식하고, 물체 중심 좌표를 스플라인(Spline) 기반 곡선형 트랙의 노드로 변환함으로써 환경에 따라 형태와 난이도가 달라지는 3차원 트랙을 구성한다. 또한 Photon Unity Networking과 World Locking Tools를 결합하여 체크포인트와 차량 상태를 실시간 동기화하고 디바이스 간 좌표계를 정렬함으로써, 두 사용자가 동일한 AR 트랙을 공유하도록 한다. 사용자는 핸드 트래킹과 제스처 인식을 통해 차량을 조작하고, 트랙 위 아이템과 장애물과의 상호작용을 통해 경쟁적인 플레이를 수행한다. 실험을 통해 평면·물체 인식과 멀티플레이 동기화, 렌더링 성능이 실시간 AR 콘텐츠 요구 수준을 만족함을 확인하였으며, 본 연구는 현실 물체를 게임 요소로 적극 활용하는 환경 적응형 AR 레이싱 시스템의 가능성을 제시하고, 향후 교육·스포츠형 콘텐츠 및 협업형 AR 응용으로의 확장을 기대한다.

국방혁신 4.0의 도입과 함께, 대한민국은 첨단 기술을 군사 훈련에 적극적으로 통합하고 있다. 현재 대한민국 군에서 사용되는 사격 훈련 방식은 프로젝터 기반 가상현실(VR) 시스템 또는 비시스루 비디오(Non Video See-Through) 방식의 HMD를 활용하는데, 이 방식은 더 많은 가상 콘텐츠 자원 (아바타 및 환경)이 필요하며 그에 따른 시스템 지원 사항(렌더링 등)이 필요하다.. 이러한 문제를 해결하기 위해 본 연구에서는 증강현실(AR) 기반 사격 훈련 시스템을 제안하며, Hololens 2와 VIVE 트래커를 활용하여 실제 총기의 위치와 신체 움직임을 정밀하게 추적하는 방법을 구현하였다. 추가적으로 Proximal Policy Optimization(PPO) 알고리즘 기반의 강화학습 AI 어시스턴트를 도입하여, 사격의 정확도와 정밀도를 분석하고 실시간 피드백을 제공하여 사용자의 자세를 개선하도록 설계하였다. 또한, TCP 및 UDP 기반 소켓 통신 방식을 채택하여 평균 2.30ms의 낮은 지연시간을 달성하며 실시간 훈련이 가능함을 검증하였다.

We propose a novel method conceptualized from the properties of physics where in particular the shape of a flame is determined by temperature that enables a control mechanism for the intuitive shaping of a flame. We focused on a trade-off issue from computer graphics whereby the turbulent flow that expresses the characteristics of the flame has a tendency to shift continuously, whereas the velocity constraints that contain a fluid within a target shape have a tendency to force movement in a particular direction. Trade-off made it difficult for animation designers to maintain a flame within the intended target shape. This paper resolves the issue by enabling the flame to be controlled without any velocity constraints by using the following two techniques: First, we model the temperature and force of the explosion generated by the combustion of explosive gaseous fuel and apply it to certain regions. Second, we expand the space of the interface between the fuel and the burned products, classifying that space into four regions and controlling the target shape of the flame by delicate adjustments to the temperature in each region. Experiments show that the flame maintains the appearance of dynamic movement while preserving the detailed 3D shapes specified by the scene designers.