Earticle

동시적 위치 추정 및 지도 작성(Simultaneous Localization And Mapping, SLAM)은 주변 환경의 지도를 구 축하며 현재 위치를 추정하는 기술로, 자율주행 로봇, 드론, 증강현실(AR) 등 다양한 분야에 활용된다. 그러나 센 서 종류와 환경에 따라 성능이 저하될 수 있는데, 예를 들어 비전 기반 SLAM은 질감이 부족한 환경에서, 라이다 기반 SLAM은 구조물이 드문 공간에서 한계를 가진다. 본 연구에서는 이러한 센서의 한계를 보완하기 위해 비전과 라이다를 융합하고, 엔코더 기반의 오도메트리(odometry)를 결합한 SLAM 기법을 제안한다. 또한, 명시적인 보정 (calibration) 과정 없이 호모그래피(homography)를 적용하여 각 센서의 좌표계를 정렬하는 방법도 함께 제안한 다. 실내외 다양한 장애물이 존재하는 환경에서 실험한 결과, 평균 위치 오차는 약 3.7cm로 측정되어 효과적으로 잘 작동함을 확인하였다.

Simultaneous Localization and Mapping(SLAM) is a technique that builds a map of the surrounding environment while estimating the current position, and it is widely used in various fields such as autonomous driving, drones, and augmented reality(AR). However, its performance can degrade depending on the type of sensor and the environment. For example, vision-based SLAM struggles in textureless environments, while LiDAR-based SLAM has limitations in areas lacking sufficient structural features. To overcome these limitations, this study proposes a SLAM method that fuses visual and LiDAR data and integrates encoder-based odometry. In addition, a homography-based approach is introduced to align the coordinate systems of each sensor without an explicit calibration process. Experiments conducted in various indoor and outdoor environments with obstacles confirmed that the proposed method operates effectively, achieving an average localization error of approximately 3.7cm.

본 연구는 자율주행 및 로봇 시스템에서의 시각적 장소 인식(Visual Place Recognition, VPR) 성능을 개선하기 위해 열적외선(Thermal Infrared, TIR) 데이터를 융합하는 새로운 방법론을 제 안한다. 제안된 방법론은 Late Fusion 방식을 채택하여, MobileNetV3를 기반으로 각 센서로부터 특징을 추출한 후, MixVPR 네트워크를 통해 두 데이터 정보를 결합한다. 실험 결과, 제안된 방법론은 기존 방법론들과 비교하여 정량, 정성적인 평가에서 우수한 성능을 보였으며, 특히 조도 변화에 따른 성능 저하를 효과적으로 극복할 수 있음을 검증하였다. 이러한 성과는 자율 주행 차량과 로봇 시스템에서 동시적 위치 추정 및 지도 작성(SLAM) 기술의 신뢰성을 높이는 데 기여할 수 있을 것으로 기대된다.

This study proposes a novel approach to enhance Visual Place Recognition (VPR) performance in autonomous driving and robotic systems by integrating Thermal Infrared (TIR) data. The proposed method employs a Late Fusion strategy, extracting features from each sensor using a MobileNetV3- based network and then combining features through the MixVPR network. Experimental results demonstrate that the proposed method outperforms existing approaches in both quantitative and qualitative evaluations, effectively mitigating performance degradation due to the illuminance variations. These findings are expected to contribute to improving the reliability of Simultaneous Localization and Mapping (SLAM) technology in autonomous vehicles and robotic systems.

배달 로봇을 위한 도시 환경, 특히 캠퍼스나 도시와 같은 지역에서는 많은 맞춤형 기능이 표준 도로 의미 분류를 무시 합니다. 이러한 과제를 해결하기 위해 우리 논문에서는 SOD(Salient Object Protection)를 활용하여 이러한 고유한 기능을 추출하고 이를 향상된 로봇 루프 폐쇄 및 위치 파악을 위한 중추 요소로 사용하는 방법을 소개합니다. 기존의 기하학적 특징 기반 위치 파악은 조명 및 모양 변화의 변동으로 인해 방해를 받습니다. 의미론적 분할보다 SOD를 선호하는 것은 표준화되지 않은 수많은 도시 특징을 분류하는 복잡성을 피하는 것입니다. 일관된 지반 특징을 달성하기 위해 MC-IPM(Motion Compensate IPM) 기술이 구현되어 왜곡 보상을 위한 모션을 활용하고 이어서 모멘트 계산을 통해 가장 관련성이 높은 두드러진 지반 특징을 선택합니다. 철저한 평가를 위해 실제 도시 시나리오에 대한 돌출성 탐지 및 위치 파악 성능을 검증했습니다.

This paper proposes an enhanced Simultaneous Localization and Mapping (SLAM) algorithm based on matching laser image and Extended Kalman Filter (EKF). In general, laser information is one of the most efficient data for localization of mobile robots and is more accurate than encoder data. For localization of a mobile robot, moving distance information of a robot is often obtained by encoders and distance information from the robot to landmarks is estimated by various sensors. Though encoder has high resolution, it is difficult to estimate current position of a robot precisely because of encoder error caused by slip and backlash of wheels. In this paper, the position and angle of the robot are estimated by comparing laser images obtained from laser scanner with high accuracy. In addition, Speeded Up Robust Features (SURF) is used for extracting feature points at previous laser image and current laser image by comparing feature points. As a result, the moving distance and heading angle are obtained based on information of available points. The experimental results using the proposed laser slam algorithm show effectiveness for the SLAM of robot.

Not all line or point features capable of being extracted by sonar sensors from cluttered home environments are useful for simultaneous localization and mapping (SLAM) due to their ambiguity because it is difficult to determine the correspondence of line or point features with previously registered feature. Confused line and point features in cluttered environments leads to poor SLAM performance. We introduce a sonar feature structure suitable for a cluttered environment and the extended Kalman filter (EKF)-based SLAM scheme. The reliable line feature is expressed by its end points and engaged togather in EKF SLAM to overcome the geometric limits and maintain the map consistency. Experimental results demonstrate the validity and robustness of the proposed method.

일반적으로 자율 주행 시스템의 구현을 위해 SLAM알고리즘이 핵심적으로 사용되어 왔다. 그러나 SLAM을 활용한 기존의 자율 주행 시스템은 정확한 지도를 생성하는 과정이 어려우며, 변동성이 높은 환경에서 취약함을 보인다. 본 연구는 SLAM에 의존하지 않는 자율 주행 로봇의 네비게이션 시스템을 구현하기 위해 Twin Delayed Deep Deterministic Policy Gradient(TD3) 강화학습 알고리즘을 활용하였다. 이 연구의 목적은 로봇이 미지의 환경에서 장애물을 효과적으로 회피하며 목적지까지 안정적으로 주행할 수 있도록 하는 것이었다. 효과적인 시뮬레이션 환경을 Gazebo를 이용하여 진행하고, 훈련된 에이전트를 직접 구현한 로봇 플랫폼에 적용하여 주행 성능을 검증하였다.

In autonomous driving, environmental perception and decision-making are technologies that acquire signals based on various sensors and generate information that enables obstacle avoidance, emergency stops, and path planning. Such environmental perception technologies are limited in that they depend on expensive sensors such as LIDAR and RADAR, so research on environmental perception technologies using only cameras is actively being conducted. Studies predicting traffic accidents based on dashcam footage have also been performed as part of such research. This is challenging because only limited forward-looking footage can be acquired, and the surrounding environment is dynamic and changes quickly, making analysis difficult. Existing research has focused on learning the spatialtemporal feature representation to solve these problems. This paper extracts the trajectories of the ego-vehicle and surrounding vehicles using Visual SLAM and Multiple Object Tracking algorithms and uses them as inputs to the graph convolutional neural network(GCN) to learn the spatialtemporal feature representation. In addition, the features learned through the GCN are used as inputs to a bayesian neural network(BNN) to predict the probability of accidents, and its ability to predict accidents in advance has been verified by comparison with existing studies.

양자 컴퓨팅의 연산 능력 발전으로 기존 공개키 암호 체계의 보안 위협이 현실화됨에 따라, 본 논문에서는 단일 수학적 난 제의 잠재적 취약점을 보완하기 위한 시스템 레벨의 'Triple-Hybrid PQC(Project Grand Slam)' 아키텍처를 제안한다. 제안 시스템은 격자(lattice), 코드(code), 해시(hash) 기반의 이종 암호 알고리즘을 논리적으로 결합하여 다층 방어 체계를 구축하였 으나, 약 21KB에 달하는 페이로드로 인해 기존 TCP 환경에서 HOL Blocking 및 전송 지연이 발생한다. 이를 해결하기 위해 본 연구는 AI 기반의 적응형 QUIC 전송 기술을 도입하였다. 경량화된 Random Forest 모델이 네트워크 상태와 데이터 민감 도를 실시간으로 분석하여, 위협 감지 시 즉시 UDP 기반의 PQC-QUIC 터널로 프로토콜을 동적으로 전환한다. 실험 결과, 제 안 시스템은 10% 패킷 손실 환경에서도 기존 TCP 대비 약 2.1배의 전송 생존성(throughput survivability)을 확보하였으며, 95.8%의 위험 탐지 정확도를 기록하여 자원 제약 환경에서의 보안성과 가용성을 동시에 입증하였다.

With the advancement of quantum computing capabilities, security threats to existing public key cryptosystems (RSA, ECC) are becoming a reality. To address the potential 'Single Point of Failure' in current Lattice-based PQC standards, this paper proposes a system-level 'Triple-Hybrid PQC' architecture, dubbed Project Grand Slam. The proposed system logically combines heterogeneous cryptographic algorithms based on Lattice, Code, and Hash problems to construct a multi-layered defense mechanism. However, the large payload of approximately 21KB causes Head-of-Line (HOL) Blocking and transmission delays in legacy TCP environments. To resolve this, we introduce an AI-based adaptive QUIC transmission technology. A lightweight Random Forest model analyzes network conditions and data sensitivity in real-time, dynamically switching the protocol to a UDP-based PQC-QUIC tunnel upon detecting anomalies. Experimental results demonstrate that the proposed system achieves 2.1 times (2.1x) higher throughput survivability compared to legacy TCP, even under a 10% packet loss environment, and records a threat detection accuracy of 95.8%. This validates that the system ensures both rigorous security and high availability in resource-constrained edge environments.

This paper shows how effectively sonar data can be worked with approaches suggested for the indoor SLAM (Simultaneous Localization And Mapping). A sonar sensor occasionally provides wrong distance range due to the wide beam width and the specular reflection phenomenon. To overcome weak points enough to use for the SLAM, several approaches are proposed. First, distance ranges acquired from the same object have been stored by using the FPA (Footprint-association) model, which associates two sonar footprints into a hypothesized circle frame. Using the Least Squares method, a line feature is extracted from the data stored through the FPA model. By using raw sonar data together with the extracted features as observations, the visibility for landmarks can be improved, and the SLAM performance can be stabilized. Additionally, the SP (Symmetries and Perturbations) model, a representation of uncertain geometric information that combines the probability theory and the theory of symmetries, is applied in this paper. The proposed methods have been tested in a real home environment with a mobile robot.

Quiet environment is essential for improving quality of living and rest. For example, noisy neighbors degrade quality of sleeping of residents. Due to emphasis on energy saving of buildings, more insulation and tighter sealing is applied for windows, doors, and walls of new buildings. As results, not only energy dissipation of new building is reduced, but also interior of new buildings have become significantly quieter. However, sound and heat insulation performance of existing buildings is relatively worse, when compared to that of recent buildings. As elapse of time, gap between bricks or concrete of building and sealing grows larger, worsening of sound and heat insulation performance of buildings is unavoidable. To provide quiet room for quality control of products or other purposes in noisy environment, indoor noise chamber is required in practice. In present work, sound transmission through indoor noise chamber is measured using different noise sources. For the worst case, noise source and sound transmission of door slam noise is shown.

In the field of autonomous navigation control of unmanned vehicles, it is most essential to estimate the location of the controlled vehicle. In this paper, a method for estimating the location of a quadcopter is proposed to autonomously control the flight of an unmanned quadcopter. The algorithm for estimating the location of a quadcopter is proposed by applying the EKF-SLAM, and the algorithm is simulated for the system to estimate and control the flight trajectories of the quadcopter. Prior to simulation, a dynamic model for the quadcopter is introduced, and simulations are carried out to determine the manner in which the quadcopter changes with respect to the thrust. Moreover, the leveling control and posture control are simulated through a PD controller, and the flight trajectories of the quadcopter are analyzed through the posture control. Furthermore, a virtual landmark is established for applying the EKF-SLAM based on a fabricated quadcopter simulator. Moreover, an observation model for observing the landmark is proposed for estimating the flight trajectories of the quadcopter, along with a state equation and a Jacobian matrix. Finally, potential advancement in the simultaneous location estimation and mapping technology of the quadcopter is presented by simulating the quadcopter EKF-SLAM and validating its performance.