고하중 작업용 상지 착용로봇 성능 평가를 위한 더미 로봇

이선휘 세종대학교 대학원 2025 국내석사

착용로봇 개발이 활발해짐에 따라 이들의 안전과 성능에 대한 검증 수요가 증가하고 있다. 하지만 기존의 착용로봇 성능평가는 착용자의 근전도 신호, 신진대사 비용과 같은 생체신호에 의존하기 때문에 착용자의 피로도, 숙련도에 따라 반복성과 객관성이 부족하다는 단점이 있다. 이를 해결하기 위해 더미 로봇을 이용한 착용로봇 성능평가 방법이 제안된 바 있다. 착용자 대신 더미 로봇이 착용로봇을 착용하고, 착용 유무에 따른 더미 로봇의 소모에너지 차이를 이용하여 착용로봇의 성능을 정량적으로 평가하는 방법이다. 따라서 고하중을 보조하기 위한 상지 착용로봇의 성능평가를 위해 새로운 상지 더미 로봇을 개발하였다. 성능평가 지표로서 소모에너지의 차이를 이용하려면 착용로봇 보조 수준에 따른 소모에너지가 선형성을 가져야 한다. 이를 위해 마찰 식별과 마찰 보정을 통하여 소모에너지의 선형성을 확보하였고, 시뮬레이션 결과 비교를 통해 소모에너지 값의 타당성을 확인하였다. 더불어 더미 로봇의 제어 정확도를 확보하여 착용로봇 성능평 가 플랫폼으로서의 필수 조건을 만족하였다. 새로운 상지 더미 로봇을 이용한 착용로봇 성능평가 방법을 실험적으로 검증하기 위해 상용 착용로봇을 적용하여 성능을 평가하였다. 실험 결과를 통해 착용로봇의 질량이 비보행/보행 시의 보조성능에 어떤 영향을 줄 수 있는지 분석하였고, 착용로봇 보조성능 설계값과 실험값을 비교하였다. 이로써 새로운 상지 더미 로봇을 이용한 착용로봇의 성능평가의 가능성을 검증하였다. As the development of wearable robots becomes more active, the demand for verifying their safety and performance has also increased. However, conventional performance evaluation methods rely heavily on the wearer’s biosignals, such as EMG and metabolic cost, which suffer from limited repeatability and objectivity due to user fatigue and proficiency. To address this issue, a performance evaluation method using a dummy robot has been proposed. In this method, the dummy robot wears the wearable robot instead of a human, and the difference in energy consumption with and without assistance is used to quantitatively assess performance. Therefore, a new upper-limb dummy robot has been developed to evaluate the performance of wearable robots designed to assist with heavy loads. For the energy consumption to serve as a valid performance metric, it must show linearity according to the level of assistance. This was achieved through friction identification and compensation, ensuring linear energy behavior. The validity of the energy consumption values was verified through simulation comparisons. In addition, the control accuracy of the dummy robot was secured, meeting the essential requirements as a performance evaluation platform. To experimentally validate the proposed evaluation method, a commercial wearable robot was applied. The results analyzed how mass of the wearable robot affects assistance performance during walking and non-walking conditions and the designed assistance performance values were compared with the experimental results. These results demonstrate the feasibility of performance evaluation using the new upper-limb dummy robot.

Novel multi-DOF counterbalance mechanism design based on a spring balancer for robot arms

이원범 Graduate School, Korea University 2022 국내박사

For many decades, robot arms have been traditionally used in the industrial fields for factory automation. To expand the applications of robot arms to indoor service sites, service robot arms have also been developed recently. Most of these service robots are equipped with a vertical prismatic joint between the mobile platform and the robot arm to allow the robot to perform various tasks. These robot arms and prismatic joints require significantly higher motor torques and power than those required by the wheel drive of a mobile platform. Therefore, the operating time of a battery-powered mobile service robot is short, and the payload capacity of a service robot arm is also low with limited task variation. A counterbalance mechanism (CBM), which is a mechanical device that cancels or reduces the gravitational torque applied to a revolute linkage joint with counterweight or springs, can be an effective solution to solve this problem. The CBM can effectively support the robot arm mass and payload, thereby reducing the motor torque and power consumption required to operate the robot. In addition, a spring balancer, which is a mechanical component that applies a constant restoring force to a wire, can be easily used to counterbalance a prismatic joint. However, mounting both the prismatic CBM and revolute CBMs on the service robot arm would substantially increase the overall volume, weight, and mechanical complexity. Therefore, a CBM that can counterbalance the arm with both prismatic and revolute joints, with a simple structure, is required. In this report, a novel multi-DOF CBM that can counterbalance a robot with prismatic–pitch–pitch joints using only one spring balancer mounted on the base is proposed. The proposed CBM consists of mechanical components such as a spring balancer, wire, and idlers. It is difficult to counterbalance the multi-DOF joints by simply connecting each joint to a single coil spring and wire, because the compensation torque of one joint can be affected by the movement of the other joints, thereby changing the spring force. Therefore, in this study, the property of the spring balancer, i.e., the restoring force is constant and not affected by a change in displacement, was exploited. The wire subjected to a constant restoring force, generated by the spring balancer initially, extends through the idlers on the link of the prismatic joint, and then forms multiple loops between the idlers, fixed to each pitch joint, and the link. Consequently, the combined force applied to each idler exerts an appropriate compensation force/ torque on each joint. To verify the performance and practical implementation of the mechanism, a 3-DOF CBM prototype along with a 7-DOF robot arm equipped with the CBM were constructed. Various experiments were conducted, and the corresponding results were compared to confirm the performance of the proposed CBM and counterbalance robot arm. It was shown that the proposed multi-DOF CBM, based on a spring balancer, properly counterbalanced the prismatic–pitch–pitch joints and enabled the construction of the high-payload service robot arm with compact actuators and link structures.

동적 환경에서 분산 로봇 경로 찾기 알고리즘

김학수 경희대학교 2010 국내석사

로봇공학은 한 로봇만이 아닌 여러 로봇과 장치들이 협력, 협동을 통하여 한계 이상의 일을 처리하는 형태로 발전하고 있다. 이러한 협력과 협동을 위해서는 정보 및 명령을 주고받는 통신능력과 위치 이동 능력이 필요하다. 본 연구는 이러한 많은 수의 로봇으로 이루어진 Networked Robots 시스템에서 발생한 정보를 처리하고, 명령을 내려 목적지까지 이동하는 분산 로봇 경로 찾기 알고리즘을 제안한다. 모든 정보를 이용하는 중앙 집중 방식은 최적경로를 찾을 수 있지만 두 가지 문제를 가지고 있다. 첫 번째, 확장성이 떨어지게 된다. 한 개의 로봇이 목적지까지 경로를 계산함으로 인하여 많은 계산량을 가지게 되며, 환경이 커짐에 따라 경우 계산량은 계속 증가한다. 두 번째, 통신 사용량이 증가한다. 국부적인 환경 변화에 시작로봇까지 정보를 전송하기 위해 메시지를 전송함으로서 전체적인 통신 사용량이 증가한다. 본 논문은 동적으로 변화하는 환경에서 확장성과 낮은 통신 사용량을 가지는 분산로봇 경로 찾기 알고리즘을 제안한다. In recent years, robotics has developed over a cooperation and coordination with many robots. Robots needs communication skills and the ability to move the location for Cooperation and collaboration. In this paper presents a networked robots system and decentralized path planning algorithm for multiple robots in large areas. The existing methods for path planning for multiple robots has two solution. In the centralized scheme can find best way to destination. but, Have two problems. First, Inflexible and Not suitable for environments changes or uncertainties. Because one robot computes a path to destination, Their complexity is high and exponential increase in large space. Communication is similar to that of the scalability. Second, Increase communicate amounts for compute a path. Decentralized methods using a localization message. Therefore, Decrease communication amount. In this paper we consider the problem of finding a best way for multiple robots in dynamic environments. We propose decentralized path planning algorithm. Decentralized methods is able to decrease communication amounts and low complexity.

Continuum Flexible Robot with Multiple Curvature : 여러 곡률을 갖는 연속체로 이루어진 휘어지는 로봇

박성호 DGIST 2016 국내석사

The continuum flexible robots are used at various fields due to characteristics of relatively more flexible than conventional rigid link-joint robots. Among many applications, this paper researches of the continuum flexible robot which can be apply with ballooning orbital floor blowout surgery. When the balloon support the fracture under the fractured orbital floor, the completeness of the surgery grows though this procedure can be performed by highly trained few surgeons. The core problem is the tip of the robot should bend most to get into the side pathway in narrow space by 1-DOF. The conventional flexible robots suppose it has equal curvature in entire flexible part or are got restrict of their structures for constant curvatures. For supplementing those problems, this paper designed the continuum flexible robot which has larger curvatures at the tip without increasing the number of actuators and analyzed the curvature ratio, the curvature according to the length of pulled wires and kinematics. The continuum flexible robot was tested by pulled wire from 0mm to 15mm and then was compared with measured value and estimated value. 연속체로 된 휘어지는 로봇들은 기존의 강체로 이루어진 로봇에 비해 유연한 특성을 갖기 때문에 여러 가지 분야에서 사용된다. 많은 적용분야 중 본 논문은 ballooning 안와하골절 수술 분야에 적용될 수 있는 continuum flexible robot에 대하여 연구했다. 안와하골절은 매우 빈번하게 일어나는 외상으로써, 골절부위를 풍선으로 받쳐주면 수술의 완성도를 높일 수 있지만 숙달된 소수의 의사만 집도 가능하다는 단점이 있다. 가장 핵심적인 문제는 로봇이 한 자유도 방향으로 움직일 때 로봇의 끝 부분이 급격하게 구부러져야 한다는 것이다. 기존의 연속체로 된 휘어지는 로봇들은 가정 또는 기구적인 구속을 통해 전체적으로 일정한 곡률을 갖도록 만들거나 와이어로 연결된 엑추에이터들의 숫자를 늘서 자유도를 증가시키는 접근법을 보여주었다. 이러한 단점들을 보완하기 위해 본 논문에서는 엑추에이터의 숫자를 늘리지 않으면서 끝 부분의 곡률이 더 클 수 있는 연속체로 된 휘어지는 로봇을 설계하고 사용한 스프링 상수에 따른 곡률 비, 당겨진 와이어 길이에 따른 곡률, 기구학 등을 분석했다. 만들어진 로봇의 와이어를 0mm~15mm 만큼 당겨서 곡률 비와 부분별 곡률을 측정하고 분석하여 나온 식과 대응 값을 적용시킨 시뮬레이션을 통해 성능평가를 실시하였다.

Human Computer Interface에 의한 서비스로봇 제어 시스템 구현

강성률 順天大學校 大學院 2005 국내석사

Present age the allotted span is increasing as becoming aging society gradually by scientific development and disabled person who have native, acquired drag as long as is in mechanization of life culture is increasing. Therefore, necessity is risen to robot about cloth and research about service robot that have Mobile Robot's form is gone a buzz. String service robot designed, and had been designed according to various kinds objective and environment that want that user is possible so that the service may help human's job to most human directly or indirectly. We service robot greatly service target is human by two type according to class. In research service robot of Human Computer Interface (HCI) via disabled person research direction catch wish to. HCI's service robot satisfies appetite which lineage disabled person wishes to act as service robot for disabled person, and it is big purpose that serve to watch by real time and prevent from danger. Control motor to sell robot necessary to take and lift CAN, PIT disease, disease etc. in refirigerator by method to satisfy this action appetite. Apply speech recognition for Interface in research. When disabled person or the two bridges that have obstacle of eight through speech recognition wishes to have object that uncomfortable disabled person wants, control motor to sell robot by command by passive character. This study used method to measure distance of reflex because do this putting out a fire of image processing to measure object distance in system that see distance of object though embodied system for measurement and speech recognition with Stereo camera during way that control service robot, and confirmed effectiveness about speech recognition because using HMM looking for characteristic of each voice using VQ after do to do qvantification using LPC by process fur speech recognition. There was error medicine 0.08% that do as result that measure distance, and speech recognition could get awareness about 80%. In this study reflex and voice method to be done and embody generalized system integration Tuesday in PC environment using camera and microphone for computer that is used much forward actually and heighten the awareness rate using service robot characteristic that improve more although embodied system that do processing individually must continue research consider.

촉각 센서를 이용한 물리적인 사람 : 로봇 정보전달 인터페이스 개발

김용호 경희대학교 2012 국내석사

로봇의 기계적, 기능적 발전으로 인해 기존 산업용으로만 쓰였던 로봇은 점차 가정, 사무실 등 다양한 분야로 확장되고 있다. 또한 로봇의 안정성 및 신뢰성이 높아짐에 따라 사람과 로봇이 같은 작업 환경에서 임무를 수행할 수 있게 되었다. 이러한 환경이 구성되면서 HRI(Human-Robot Interaction) 기술에 대한 관심도가 높아지고 있다. PHRI(Physical Human-Robot Interaction)은 HRI 기술 분야 중 하나로 사람과 로봇의 물리적인 상호 정보 전달을 위한 기술 개발 및 연구를 하는 학문이다. PHRI 학문은 크게 물리적인 상호작용을 위해 시스템을 개발하는 분야와 물리적 센서를 개발하는 분야로 나뉜다. 그러나 이 두 분야의 연구가 개별적으로 진행되어 학문적 성과를 크게 내지 못하고 있는 상황이다. 본 논문에서는 이러한 문제점을 극복하기 위해 촉각 센서 중 하나인 터치 센서를 이용하여 사람과 로봇이 물리적 정보 전달을 할 수 있는 인터페이스 시스템을 개발하였다. 본 시스템을 통해 물리적인 정보를 획득할 수 있으며, 이를 적절히 가공하여 다른 시스템에 획득한 정보를 전달하게 된다. 또한 시스템을 개발자가 쉽게 입력 정보를 얻기 위해 인터페이 스 시스템에 API(Application Program Interface)를 정의하고 구현하였다. 본 논문에서 제안한 인터페이스 시스템의 검증을 위해 모바일 로봇의 원격 제어 시스템을 구성하고 인터페이스로 하여금 물리적 정보 획득이 가능하도록 하였다. 실험 결과 터치 센서로부터 입력된 정보가 알맞게 가공되어 로봇 제어기로 전달되고, 모바일 로봇이 사용자가 원하는 동작을 수행함을 확인하였다. Robot has been expended from industry to home and office according to mechanically and functionally advancement of robotics. Also improvement of safety and reliability of robotics allows robot can works with human in same environment. In this situation, HRI(Human-Robot Interaction) technology is issued and got spotlight. PHRI(Physical Human-Robot Interaction) is a sub technology of HRI and has studying how to interact between human and robot using physical sensors, what is the best design of robotic system for physical human-robot interaction. PHRI is divided into two parts. One is research for physical sensors, and the other is system design for PHRI. However, both parts have not tried to combine sensor and system except some special cases. On this situation, academic results are not anticipated without combining two parts. In this paper, interface system that can interact with human using tactile sensor is proposed to resolve the problem. Through proposed system, getting physical information from touch sensor is possible. Acquired information is processed to generate input vector for providing it to a user or developer, system designer. Proposed interface system also provides API(Application Program Interface)s to supply their own system.To validate proposed system, remote control system for mobile robot is constructed and used. Remote control system acquires input data which is made by proposed system using touch panel. As the experiment result, input data is well-generated through input vector generation module in proposed system and is transmitted to remote control system to drive mobile robot. Finally, mobile robot drives to specific direction depending on input vector.

초등학교 로봇교육을 위한 로봇 하드웨어 구성 및 인터페이스의 제안

권갑순 인천대학교 정보통신대학원 2009 국내석사

Robot Controllers, Gaze Behaviors and Human Motion Datasets for Object Handovers

Kshirsagar, Alap ProQuest Dissertations & Theses Cornell University 2022 해외박사(DDOD)

We investigate the collaborative task of object handovers between a human and a robot. Object handover is one of the most common skills required for a collaborative or an assistive robot. Tasks such as surgical assistance, housekeeping, rehabilitation assistance, and collaborative assembly require a robot to give objects to a human (robot-to-human handover) and take objects from a human (human-to-robot handover).First, we design robot controllers for previously unexplored human-robot handover scenarios involving a) robot behavior specification by end-users, b) flexibility to change handover strategies, c) unknown robot dynamics, and d) human-like bi-manual handovers. For "a", we propose two receding horizon controllers that utilize timing parameters and provide failure feedback to the user. We find that these two controllers offer contrasting user experience and task performance, vis-a-vis a baseline controller, in an industrial human-robot collaborative task. For "b", we present a controller that uses automated synthesis from specifications in Signal Temporal Logic, and illustrate the flexibility of our approach by reproducing, in a simulation environment, existing human-robot handover strategies found in the literature. For "c", we evaluate the potential of a reinforcement learning method, Guided Policy Search (GPS), both in a simulation environment and on a physical robot arm. Our evaluations provide new insights into the capabilities and limitations of GPS. For "d", we evaluate the potential of an imitation learning technique, Bayesian Interaction Primitives (BIPs), and discuss two adaptations of BIPs for generating humanlike bimanual reaching motions.Second, we investigate robot gazes in human-to-robot handovers. We design human-inspired robot gaze behaviors by analyzing the receiver's gaze patterns in human-to-human handovers. We conduct four user studies with two robot platforms to evaluate human preferences for the robot receiver's gaze behavior in handovers. Our studies reveal that the most preferred robot receiver's gaze behavior in a human-to-robot handover consists of a combination of face-oriented gaze for social engagement and task-oriented gaze directed at the giver's hand.Third, we develop two datasets of human-to-human handovers: bimanual handovers, and multiple sequential handovers in a shelving task. These datasets could help build human motion models and robot controllers for those scenarios of human-robot handovers.

Enhanced Human-Robot-Environment Interactions Enabled by Multi-Stiffness Soft Sensors and Interactive Control

박명선 서울대학교 대학원 2024 국내박사

This thesis addresses the grand challenge of making the robots friendly to humans and adaptive to the surrounding environment. The thesis makes attempts to improve the human-robot-environment interaction loop by seeking solutions from soft robotics technologies. To achieve the ultimate goal, three practical problems are defined. The first problem is to accurately perceive each of the human, robot, and the environment. In Chapter 2, the multi-stiffness sensor structure with improved sensitivity is proposed to solve the problem. Theoretical modeling and experimental validation are presented to prove the performance of the structure. The second problem is to measure the motion of human hands, which are actively and elaborately used for interacting with the robots and environments. To achieve both the accuracy of sensing and the structural simplicity, in Chapter 3, the multi-stiffness sensor-based wearable glove is developed and the post-processing method for collecting accurate ground truth data for calibration of the glove is proposed. The hand-tracking system is able to achieve unprecedented accuracy and robustness compared to the previous studies, which are shown through quantitative and qualitative evaluations. The third problem is to connect the human and soft robots, by implementing self-sensing functionality to a soft robot for accurate control of its motion and developing a haptic device to transmit back the robot’s states to the human. In Chapter 4, an ionic electroactive polymer-based actuator is selected as a solution because of its structural compactness, low input voltage, and inherent safety. By integrating the soft sensor into the actuator and using the customized haptic device, accurate teleoperation of the soft gripper system for a practical pick-and-place task is demonstrated. The last problem is to connect the soft robots and the surrounding environments to complete the entire interaction loop. In Chapter 5, a method of hybrid system analysis is proposed that allows the high complexity of the soft robots’ kinematics to be simplified. The method enables the robot to identify and adapt to the environment using the embedded sensors while interacting with it. A pneumatic gripper system trying to grasp an unknown object is selected as an example to implement the method. Based on the method, the robotic gripper successfully completes the given task by estimating the unknown size of the object and modulating its body position to grasp the object. Keyword : Robotics, Human-Robot Interaction, Robot-Environment Interaction, Soft Robotics, Grippers, Multi-material, Human Motion Tracking, Hybrid System. 본 논문은 인간 친화적이고 주변 환경에 적응할 수 있는 로봇 시스템의 개발을 가장 큰 목표로 다룬다. 논문은 인간, 로봇, 환경 상호작용의 성능의 개선을 위해 소프트 로봇 기술로부터 그 해결책을 찾고자 한다. 궁극적인 목표의 달성을 위해 세가지의 실질적 문제가 정의된다. 첫번째 문제는 인간, 로봇, 환경 각각의 시스템을 정확하게 인식하는 것이다. 2장에서는 이 문제를 해결하기 위해 개선된 민감도를 갖는 다중 강성 센서 구조를 제안한다. 그리고 이 구조의 성능을 이론적 모델링과 실험적 검증을 통해 입증한다. 두번째 문제는 로봇 및 환경과의 상호작용을 위해 가장 적극적이고 정교하게 사용되는 인간 손의 움직임을 측정하는 것이다. 3장에서는 측정의 정확성과 구조의 간결함을 동시에 확보하기 위해 다중 강성 센서 기반의 착용형 장갑을 개발하고, 이 장갑의 캘리브레이션을 위한 정확한 레퍼런스 데이터 수집을 위한 후처리 방법을 제안한다. 또한 정량 및 정성 평가를 통해 제안한 손 트래킹 시스템의 우수한 정확성과 강건성을 입증한다. 세번째 문제는 소프트 로봇의 정확한 동작 제어를 위한 자가 감각 기능의 구현과 이 로봇의 상태를 다시 인간에게 전달할 수 있는 햅틱 디바이스 개발을 통해 인간과 로봇을 연결하는 것이다. 4장에서는 구조적 간결성, 낮은 입력 전압 및 고유한 안전성을 갖춘 이온 기반 전기 활성 고분자 구동기를 해결책으로 선택한다. 소프트 센서가 통합된 고분자 구동기와 햅틱 디바이스가 통합된 소프트 그리퍼 시스템의 정확한 원격 조작을 통한 실용적인 파지 및 놓기 작업을 시연한다. 마지막 문제는 소프트 로봇과 주변 환경을 연결하여 전체 상호작용을 완성하는 것이었다. 5장에서는 높은 복잡도를 갖는 소프트 로봇의 기구학을 단순화하기 위한 하이브리드 시스템 분석 방법을 제안한다. 이를 통해 환경과 상호작용하는 동안 로봇은 내장된 센서를 활용하여 환경을 식별하고 이에 적응할 수 있다. 이 방법을 실제로 구현하기 위해, 미지의 물체를 잡기 위한 공압 그리퍼 시스템을 예로 든다. 제안된 분석 방법을 통해 로봇 그리퍼는 미지의 물체의 크기를 추정하고 이에 기반해 자신의 위치를 조절함으로써 주어진 작업을 성공적으로 수행한다.

Object detection and semantic segmentation applied on robots

Shenlu, Jiang Sungkyunkwan university 2020 국내박사

Camera is an important sensor to enlighten the robot to perceive and understand the real-world. The convenience of the camera leads the robot vision to become a key technology for autonomous artificial robots. Differing from general computer vision task, the computing resources, facilities and operation environments for robot tasks are specific and require to be optimized with adaption to the restrictions. In this dissertation, we are interested improving object detection and image segmentations methods to accomplish robot vision systems. Currently, deep neural networks have already proved its advancing accuracy comparing with the methods based on human-crafted features. Accompanying with the accuracy booming up, the computing cost of the deep neural network is also exhaustive, which demands high-performance graphic card to operate the program. However, it is hard to be operated on the mobile robot because of the facility restriction. We propose a classification-lock strategy to combine the advantages of online classifier and deep neural network to keep both performance and detection quality. The classification-lock strategy employs the online classifier merging with the lightweight deep neural network to enable the mobile robot to rapidly detect the target object in the person following and multi-story elevator button detection task. With technology improvement, the large size ground robots (e.g., self-driving car) are able to provide enough computing power for deep learning. The semantic segmentation is widely employed in the scene understanding for self-driving car operation. The operation of self-driving car requires rapid response, but the semantic segmentation is with exhaustive computing cost. The depth-wise asymmetric bottleneck with point-wise aggregation decoder is proposed to extract the features with less parameters and enable the real-time semantic segmentation in the urban street view scene. Except ground robot tasks, robot vision in the aerial robots are also common for several geoscience tasks, and those tasks do not require online processing. However, the objects are with very different scales in such tasks. In addition, some objects are in specific shapes which are hard to be localized. We present an object detection neural network using double shot with misplaced localization strategy to optimize the object detection in the aerial view, especially, the small size and narrow rectangular object detection. One more issue is the feature adaption. The basenets trained by ground view images are hard to fine-tune the features in the aerial view. In addition, the noises from the environment greatly influence the detecting accuracy. Moreover, the hyperspectral and color infrared images are also widely used, but they are different type of images. Therefore, we propose a double pyramid encoding decoding neural network for aerial view land use semantic segmentation task on building/road damaging evaluation after earthquake, and the dead wood detection in the CIR images.