4면 매립형 광시야각 라이다를 이용한 무사각 전방향 감지 모바일 로봇 시스템

본 기술은 모바일 로봇 하단의 전·후·좌·우 4면에 매립 배치된 광시야각 라이다 센서로부터 깊이 영상을 획득하고, 센서별 캘리브레이션 파라미터를 이용해 단일 World 좌표계로 정합하여 사각지대 없는 통합 3D 포인트 클라우드를 생성하는 감지 시스템입니다.

기존 로봇 상단 탑재형 라이다는 고가이면서 부피가 크고, 센서 사각지대로 인해 로봇 하부와 구동부 근처의 장애물을 감지하지 못해 별도의 보조 센서가 필수적이라는 비효율이 있었습니다.

본 기술은 4면에 매립된 라이다로 수평 전방향 서라운드 뷰와 수직 30도 이상의 화각을 확보하고 회전변환 행렬과 원점 좌표를 이용해 개별 센서 데이터를 병합하는 방식을 제안합니다. 라스트 마일 배송 로봇과 실내 서비스 로봇에 적용될 수 있어 보조 센서 없이 음영 지역을 제거하여 주행 안전성과 원가 경쟁력을 동시에 확보합니다.

‍

‍

Key Features:

• 모바일 로봇 하단에 매립형으로 전면·후면·좌측면·우측면의 4면에 각각 라이다 센서를 탑재하는 단계
• 4면에 탑재된 각각의 라이다 센서에서 주변 환경에 대한 깊이 영상을 얻는 단계
• 각 라이다의 위치정보를 이용한 캘리브레이션을 통해 통일된 3D 좌표계로 깊이 영상을 정합하는 단계
• 회전변환 행렬과 원점 좌표를 적용하여 여러 깊이 영상을 하나의 포인트 클라우드로 병합하는 구성

‍

본 발명은 과학기술정보통신부의 광시야 고해상도 라이다 기반 라스트 마일 자율주행 로봇 플랫폼 지원을 통해 개발되었습니다.

Fault-Tolerant Reinforcement Learning for Legged Robots Using Knowledge Distillation and Joint Trajectory Space Learning

This technology is a reinforcement learning-based gait control method that enables robots to quickly resume adaptive walking when hardware failures, such as leg damage, occur. It achieves this by distilling knowledge from an agent trained in a normal state and utilizing it as a refined joint trajectory space through an encoder-decoder neural network.

Existing gait control technologies can adapt to terrain or environmental changes, but they face inefficiencies when hardware failures occur, often leading to a loss of control or requiring the agent to be retrained from scratch.

This technology proposes a method that uses a conditional variational autoencoder to set the joint trajectory space as the action space, narrowing the search space during failures based on knowledge learned in a normal state. It generates anchor points and paths based on conditional vectors to derive optimal joint trajectories in real time. This ensures robust autonomy, allowing robots to autonomously reconfigure their gait even when legs are damaged, making it ideal for environments where mission interruption is critical, such as disaster site exploration, defense, and industrial patrolling.

‍

‍

Key Features:

• Generating anchor points corresponding to conditional vectors in the gait learning method for n-legged robots
• Constructing a joint trajectory search space by generating paths corresponding to the generated anchor points
• Performing simulations for joint trajectories corresponding to the paths and obtaining rewards
• Updating the gait policy based on the obtained rewards and learning the gait policy

‍

This invention was developed with support from the Artificial Intelligence Graduate School Program (Korea University) funded by the Ministry of Science and ICT.

Robot Arm Control System Integrating Augmented Reality Object Detection and 3D Coordinate Calculation

This technology is an AR-based robot arm control system that detects objects in real-time from video captured via an AR device and calculates their 3D coordinates using ray casting and mesh generation to precisely control the target position of a robot arm.

Conventional controller or eye-tracking methods are difficult for individuals with physical disabilities, such as quadriplegia, to operate directly. Furthermore, these methods present inconveniences and collision risks, as users must simultaneously monitor the screen and the robot arm's position.

This technology proposes a method where an AR device recognizes the user's gaze to select a specific object, calculates the relative distance and 3D coordinates between the object and the robot arm, and enables the robot arm to automatically move and perform grasping tasks, ensuring intuitive and safe operation. It can be utilized for rehabilitation assistance, support for daily living for people with disabilities, and remote operations, significantly improving the independence and quality of life for users with physical limitations.

‍

‍

Key Features:

• AR device including camera, sensor, and display modules to extract distance and coordinates of objects
• Data processing unit that acquires video captured from the AR device in real-time to perform object detection
• Configuration that displays object detection results as bounding boxes on the display module
• Configuration that calculates the 3D coordinates of the selected bounding box to control the target position of the robot arm

‍

This invention was developed with support from the Ministry of Science and ICT for the "Development of Customized Brain-Robot Interface for the Physically Disabled with Improved Accuracy, Speed, and Convenience" and the "Development of Non-invasive BCI Integrated Brain-Cognitive Computing SW Platform Technology for Controlling Real-life Appliances and AR/VR Devices via Thought" (BCI-General/Sub-project 1) projects.

AI-Based Approximate Path Planning Technology Using Embedding Distance-Based Substitute Object Detection

This technology is an AI-based approximate path planning device and method that identifies substitute objects or approximate spaces when a target object is not detected by utilizing distances in an embedding vector space, and re-plans the robot's destination and movement path accordingly.

Previously, if a target object commanded by a user was not detected in the surrounding environment, it was impossible to set an endpoint, causing the robot's path planning to be interrupted and the movement task to fail.

This technology proposes a method that uses an embedding algorithm to extract a substitute object with an embedding value closest to the target object, or identifies an approximate space where the target object is highly likely to exist, and generates a path to that point. This allows tasks to continue without interruption even when recognition fails. It is applied to home service robots and indoor delivery robots to ensure autonomy that flexibly responds to the uncertainties of real-world environments.

‍

‍

Key Features:

• Receiving a command indicating a movement goal by an autonomous robot and initiating path planning
• Determining whether a target object corresponding to the command exists within the captured surrounding video images
• Extracting a substitute object with the smallest embedding distance if the target object is not identified
• Generating a movement path to the extracted substitute object or approximate space and executing the navigation

‍

This invention was developed with support from the Artificial Intelligence Graduate School Program (Korea University) funded by the Ministry of Science and ICT.

4-Module Pneumatic Multifunctional Soft Robot with Snap-Joint Shape Morphing and Retention

This technology is a multifunctional soft robot mechanism that combines four pneumatically driven modules with two snap-through joint sections, allowing the robot to change and maintain its geometric shape using only a single pneumatic control.

Conventional pneumatic network soft robots require continuous air supply to maintain their shape and demand complex inputs, which increases the overall volume and weight of the device.

This technology introduces bistable shell-structured snap joints that use snap-through and snap-back behaviors triggered by critical pressure to lock the robot's physical shape. This enables the implementation of a soft robot capable of dynamic mode switching, such as aligning the four drive modules in a line or deploying them horizontally depending on the control mode. Because it can switch between various movement modes with a single pneumatic input, it offers exceptional competitiveness in environments requiring multifunctionality with limited resources, such as exploration and disaster response robotics.

‍

‍

Key Features:

• Four drive modules featuring internal spaces that can expand or contract longitudinally via pneumatic pressure, connected in series at both ends.
• Four guide hinges that rotatably connect the joints between adjacent drive modules.
• Two snap-joint sections installed between alternating connection points, capable of snap-through behavior via pneumatic pressure.
• A control unit that individually regulates the pneumatic pressure applied to the four drive modules to change and maintain the robot's shape.

‍

This invention was developed with support from the Metamorphic Mechanical Systems Research Group of the Ministry of Science and ICT.

Variable Stiffness Joint for Soft Robots Using Pneumatic Snap-Through Buckling

This technology is a variable stiffness joint that drives mechanical structures by utilizing pneumatic snap-through buckling. It features a snap joint module positioned between two pneumatically pressurized members to deform the shape of a soft robot.

Conventional pneumatic network-based soft robots require complex, separate input controls for every movement and continuous pressure supply to maintain their deformed state.

This technology proposes a design that combines an elastic shell capable of pneumatic snap-through behavior with a tendon structure, enabling shape deformation and state retention with a single input control. This allows the deformed shape to be stably maintained without continuous pressure supply. It significantly reduces the burden on pressure supply systems in applications requiring lightweight and low-power operation, such as medical assistive devices, wearable devices, and grippers, thereby expanding the practical range of soft robots.

‍

‍

Key Features:

• A first member having a first internal space for pneumatic supply, supporting one side of the snap joint unit
• A second member having a second internal space for pneumatic supply, supporting the other side of the snap joint unit
• A snap joint unit connecting the first and second members, capable of snap-through behavior via applied pneumatic pressure
• A control unit for regulating the pneumatic pressure applied to the internal spaces of the first and second members

‍

This invention was developed with support from the Metamorphic Mechanical System Research Center of the Ministry of Science and ICT.

Quadrotor Aircraft Implementing 5 Degrees of Freedom via Single Servo Motor Mechanical Tilting

This technology is a quadrotor-based tilt-rotor aircraft. It features a mechanical mechanism that connects rotor shafts at the front and rear of the body to a single servo motor using a belt-pulley or gear transmission structure, allowing for synchronized tilting of all rotors, along with a 5-degree-of-freedom control method utilizing this mechanism.

Conventional multi-rotors are limited to 4 degrees of freedom because their thrust direction is fixed relative to the airframe. This makes translational movement impossible without tilting the entire aircraft and restricts stable hovering while in a tilted state.

This technology uses a mechanical tilting system with a single servo motor to tilt the rotation axes of all rotors simultaneously, enabling thrust direction control independent of the airframe's attitude. It proposes a controller that calculates optimal control inputs based on a dynamic model decomposed into underactuated and fully actuated subsystems, allowing for 5-degree-of-freedom flight with minimal actuators. It is highly applicable to missions where tilting the airframe is not feasible, such as precision photography, facility inspection, and close-proximity flight in confined spaces, meeting the demand for high-performance aircraft with minimal hardware.

‍

‍

Key Features:

• A first drive unit including a first rotor shaft installed horizontally at the front of the body and a pair of rotors installed vertically at both ends.
• A second drive unit including a second rotor shaft installed horizontally at the rear of the body and a pair of rotors installed vertically at both ends.
• A rotor tilting unit that transmits the driving force of a servo motor installed in the middle of the body to rotate both rotor shafts in the same direction by the same amount.
• A controller that outputs tilting control signals to the rotor tilting unit to manage the fully actuated and underactuated subsystems.

‍

This invention was developed with support from the Ministry of Science and ICT's development of image-based detection and avoidance technology, and the Ministry of Education's development of tilt-rotor control techniques based on coupling/uncoupling mechanisms for autonomous cooperative transport.

Glove-type motion recognition system measuring resistance and capacitance changes in conductive fibers

This technology analyzes gesture information by measuring electrical signal changes (resistance and capacitance) resulting from joint bending in the finger and palm areas of a glove woven with conductive fibers.

Conventional data gloves require complex manufacturing processes and incur high cutting and sewing costs due to the integration of wires, electronic sensors, and circuits.

This technology integrates the sensor area by weaving conductive fibers into the same layer as non-conductive fibers, configured to detect changes in finger joint contact points and palm capacitance. Applicable to rehabilitation training, wearable interfaces, and remote robot operation, it enhances gesture detection performance with a simple structure that eliminates the need for embedded circuits.

‍

‍

Key Features:

• A glove featuring palm and finger areas woven from non-conductive fibers, with a first sensing area woven from conductive fibers at the finger joint regions.
• A sensing pad woven to face the palm area, featuring a second conductive area woven from conductive fibers.
• A plurality of first conductive areas woven from conductive fibers on the same layer as the palm area, arranged in parallel in one direction.
• A glove-type motion recognition system that measures capacitance at each point where the first and second conductive areas intersect.

‍

This invention was developed with support from the Ministry of Science and ICT for the development of biomimetic bionic arm mechanisms.

Social robot for the hearing impaired that detects sounds via indoor autonomous navigation and provides alerts through physical contact

This technology is a robot control mechanism that autonomously navigates indoor spaces to detect sound information. When it determines an alarm situation requires the user's attention, it sends information to the user's device or, if the user does not respond, moves directly to the user to provide an alert through physical contact.

The hearing impaired often face risks or daily inconveniences due to an inability to perceive indoor sound information (such as fire alarms, doorbells, or household appliance sounds) in a timely manner, while existing attachable devices are cumbersome to install and often have blind spots.

This technology features a wheeled body equipped with sound, location, and object detection sensors, utilizing AI algorithms to analyze the user's location and sound data. Upon detecting an alarm, it sends a notification to the user's device; if unconfirmed, the robot navigates to the user's location and induces physical contact (tactile stimulation) by repeatedly moving forward and backward to ensure the alarm is perceived. Applicable to home service robots, indoor safety, and accessibility support, it enhances daily safety by delivering sound information to the hearing impaired immediately.

‍

‍

Key Features:

• An object detection sensor coupled to the body that captures the surroundings to detect the relative position and shape of nearby objects.
• A location sensor coupled to the body that performs 3D scanning of the surrounding space to detect the body's relative position within that space.
• A home social robot device for the hearing impaired that controls the display to show the type and location of detected sounds.
• A sound detection sensor coupled to the body that detects sound information occurring in the vicinity.

Modular Robot Actuator Module with Worm Gear Reduction and Connector Coupling

This technology is a modular actuator featuring a worm and worm gear reduction system. By positioning the motor and motor driver both inside and on the exterior of the case and separating them from the rotating assembly, it ensures ease of assembly and flexibility in capacity scaling. The module allows for the construction of multi-jointed robot arms by interconnecting multiple units via integrated coupling interfaces and connectors.

Previously, designing individual robots required custom manufacturing of components and frames, leading to high production costs. Furthermore, a lack of modularity in robot drive units made maintenance and performance upgrades difficult.

This technology adopts a structure where the motor is fixed to the exterior side of the case, driving the worm gear of the rotating assembly via a worm shaft. It enables the direct mechanical and electrical connection of multiple drive modules using coupling interfaces and pin connectors. Applicable to collaborative robots, logistics manipulators, and educational robot platforms, it allows for the configuration of various robot arm specifications simply by combining modules, significantly reducing development time and costs.

‍

‍

Key Features:

• A modular case featuring a rotating assembly on one side and a first coupling interface on the other for connection to external components.
• A motor mounted on the exterior side of the case, configured to transmit rotational drive force to the rotating assembly.
• A second coupling interface located at the end of the rotating assembly for connection to external components, and a worm gear that receives drive force via a worm gear shaft.
• First and second connectors provided at the first and second coupling interfaces, respectively, and electrically connected to the motor driver.

‍

This invention was developed with the support of the Ministry of Science and ICT for the development of a modular manipulator based on spherical parallel complex joints for item delivery and collection.

근전도와 이벤트 카메라를 결합한 스파이킹 뉴럴 네트워크 기반 로봇 팔 제어 기술

본 기술은 EMG 센서의 근전도 데이터와 DVS 카메라의 시각적 움직임 정보를 입력받아 적응형 필터와 델타-시그마 변조를 통해 스파이크 신호로 변환하고, 이를 멀티 스파이킹 뉴럴 네트워크에 입력하여 실시간으로 손과 팔의 동작을 분류하고 재현하는 뉴로모픽 제어 기술입니다.

종래의 근전도 및 가속도 센서 기반 제어 방식은 손과 팔의 동작을 정밀하게 모방하는 데 한계가 있었고, 실시간 반응성이 낮으며 로봇 제어 시 소비 전력이 높아 의료용 로봇 시스템의 성능 발전에 제약이 있었습니다.

본 기술은 근전도 데이터를 스파이크 신호로 변환하고 DVS 카메라 데이터를 크로핑과 다운-샘플링을 거쳐 SNN 모델에 병렬 입력함으로써 연산 효율을 높이고 저전력으로 고속의 정밀한 동작 모방 제어를 수행합니다. 의수와 재활 로봇, 원격 조작 매니퓰레이터에 적용될 수 있어 배터리 부담을 낮추면서 사용자의 의도를 즉각 반영하는 새로운 가능성을 제시합니다.

‍

‍

Key Features:

• EMG 센서로부터 측정된 사람의 손 또는 팔의 동작과 관련된 근전도 데이터를 수신하는 통신부
• DVS 카메라가 감지한 손 또는 팔의 움직임에 대한 정보를 함께 수신하도록 구성되는 통신부
• 수신된 근전도 신호를 적응형 필터로 특징 추출한 후 델타-시그마 변조를 수행하여 스파이크 신호로 변환하는 구성
• 변환된 스파이크 신호와 DVS 카메라의 움직임 정보를 학습된 멀티 스파이킹 뉴럴 네트워크 모델에 입력하는 구성

‍

본 발명은 과학기술정보통신부의 인간의 신경계를 모사한 뉴로 칩 설계 기술 및 뉴로 컴퓨팅 플랫폼 연구개발 지원을 통해 개발되었습니다.

Domain Adaptation-Based Gait State Prediction Technology with Enhanced Accuracy

This technology relates to a gait state prediction system and method using domain adaptation techniques and flexible time windows, enabling highly accurate estimation of gait state variables despite individual differences in walking patterns.

Existing gait state prediction models suffer from a sharp decline in accuracy when users or environments change, and their fixed time windows limit their ability to adapt flexibly to variations in walking speed.

By integrating domain adaptation algorithms with flexible time window techniques, this technology ensures robust predictive performance against individual differences and speed variations. It can be applied to gait rehabilitation, wearable robot control, and healthcare monitoring.

‍

‍

Key Features:

• A database that stores motion signal values and ground truth gait state variables from previous test subjects.
• A motion signal measurement unit composed of inertial sensors that measures motion signal values, such as the thigh angle and angular velocity, of new test subjects.
• A feature extraction unit that extracts the characteristic factors used for gait prediction from the motion signal values of new test subjects.
• A gait state variable prediction unit that uses stored data as source data and extracted characteristic factors as target data to perform predictions via domain adaptation techniques.

‍

This invention was developed with support from the Korea Forest Service’s project for developing deep learning-integrated smart wearable suits to assist muscle strength, prevent injuries, and improve work efficiency for forestry workers; the Ministry of Science and ICT’s Zero-Power Human Augmentation Basic Research Laboratory; and the development of deep learning-based tactile/texture analysis and tactile-feedback augmented prosthetic hands using flexible artificial neural patches.

Liquid motion detection technology for spill-free delivery robots

This technology relates to a liquid motion detection system for delivery robots and a liquid delivery robot equipped with said system. More specifically, it involves technology that analyzes the sloshing of liquid on a tray in real-time using a camera to control the robot's driving speed and tray angle.

Service robots transporting food and beverages in restaurants frequently face issues with liquid spilling due to sudden speed changes or uneven floor surfaces. This leads to reduced service quality, the need for rework, and potential safety hazards.

By actively controlling the tray angle and driving speed through a liquid state analysis unit and a reinforcement learning model, this technology ensures stable delivery without liquid spills. It can be applied to various fields, including serving robots, unmanned cafes, and hospital transport robots.

‍

‍

Key Features:

• System configuration for detecting the state of at least one liquid container placed on a delivery robot's tray during transport
• At least one camera that captures images of the liquid container's rim and the surface of the liquid stored within
• A liquid state analysis unit that detects the state of the liquid in real-time during delivery by analyzing the captured images
• Configuration that detects the rim of the liquid container using an image-based object tracking algorithm and senses the movement of the liquid within that area

‍

This invention was developed with support from the Ministry of Science and ICT’s Zero-Power Body Enhancement Basic Research Laboratory and the development of deep learning-based tactile/texture interpretation and tactile-enhanced prosthetic hands using flexible artificial neural patches.

High-Speed Data Transfer System Utilizing Thread Scheduling and Memory Buffer Placement per CPU Socket

This technology is a data transfer system that configures both the sender and receiver with multiple CPU sockets and socket-specific memory buffers, assigning master, communication, and I/O threads to each CPU socket to facilitate data transfer between nodes.

When transferring large volumes of data between nodes, bottlenecks often occur between storage, CPU, and memory, leading to reduced transfer efficiency due to memory access latency and memory controller load.

By assigning threads to each CPU socket based on memory buffer location and placing the master and communication threads within the same NUMA node, this technology reduces memory access latency and controller load, thereby increasing transfer efficiency in large-scale data environments such as high-performance computing and big data processing.

‍

‍

Key Features:

• A data transfer system comprising multiple CPU sockets and memory buffers allocated to each CPU socket.
• Each CPU socket includes CPU cores assigned to master threads, communication threads, and I/O threads.
• Each CPU socket is configured to include multiple NUMA nodes composed of the CPU cores within that socket.
• The master thread and communication thread are configured to be assigned to CPU cores within the same NUMA node.

‍

This invention was developed with support from the Ministry of Science and ICT's Core Technology Development Project for Big Data Processing Advancement.

‍

Biohybrid Robot Using Gold-Hyaluronic Acid Nanoparticle Muscle Bundles

This technology involves a biohybrid robot and a method for manufacturing muscle bundles. The robot’s drive unit consists of muscle bundles cultured with gold nanoparticles immobilized with hyaluronic acid, while its propulsion unit is composed of a plate-shaped support and an oar.

Conventional industrial robots, due to their rigid structures, have limited adaptability to their surroundings. Soft robots, intended as alternatives, have struggled to achieve the driving force and cellular bioactivity levels comparable to living muscle tissue.

By incorporating gold-hyaluronic acid nanoparticles, this technology enhances the electrical conductivity and cellular bioactivity of muscle bundles and proposes a method for connecting multiple mobile units. It can be utilized for drug screening for muscle disease prevention and treatment, as well as for the development of bio-actuators, significantly advancing the driving force and practicality of tissue-based robots.

‍

‍

Key Features:

• Propulsion unit comprising a plate-shaped support and an oar extending longitudinally from one end of the support
• Drive unit comprising a pair of pillars arranged vertically in parallel beneath the support, surrounded by muscle bundles
• Muscle bundles manufactured by co-culturing muscle cells with gold nanoparticles immobilized with hyaluronic acid
• Configuration further including a connector that links the supports of two or more parallel mobile units

‍

This invention was developed with support from the Ministry of Science and ICT’s project for brain-assembloid-based biomimetic sensing biohybrid robots, the Ministry of Science and ICT’s project for organoid-based nanobiohybrid actuator chips for drug screening, and the Ministry of Education’s project for developing nanobiochips for brain disease drug evaluation.