Keywords: Soft Robotics, Actuation and Actuators
Abstract: In this paper, we explored calculating the design parameters of the actuator to change the soft rotation actuator into the desired shape after deformation. The angle of the fiber, a component of the actuator, was set as a design parameter so that the hyperelastic elastomer approximated as a thin shell takes the shape of a double curvature shell with a specified geometric value after deformation. To this end, the equilibrium equation was derived using the metric tensor of the shell, and the general solution of stress was obtained from the equation to describe the deformation shape mathematically. The angle of the fiber was calculated mathematically with some assumptions.

Keywords: Soft Robotics, Actuation and Actuators, Mechanism and Design
Abstract: Actuators are the building block of robots as they enable their motion. Although sensors and mechatronic systems are of high importance, the actuation performance of their constituent motors or artificial muscles is what will limit the overall performance of the system and prevent them from being used in a wide range of scenarios. This paper details the design of an artificial muscle making use of both positive and negative pressures to produce a linear actuation. This artificial muscle is capable of full contractions with payloads of 80 kg, powerful actuation with an average power of 100 W, and was implemented as the bicep in a robotic arm capable of handling payloads of 5 kg.

Keywords: Soft Robotics, Dynamics and Control, Modeling, Identification, Calibration
Abstract: This paper introduces a pneumatic pressure booster and pressure controller for soft robots. The pressure booster and pressure controller were designed for the pneumatic power source of soft robots. The pressure booster supplies a large flow rate instantaneously by opening an air tank connected to a pneumatic circuit that generates pressure with a pneumatic pump. The pressure controller simultaneously controls the positive and negative pressure using the inlet and outlet of a single pneumatic pump. Two valves that connect the inlet and output port to the atmosphere were controlled by PID control and disturbance observer loop. It was confirmed that the time of reaching the target pressure of 100 kPa decreased by 1.15 seconds and the supplied flow rate increased by up to twice compared to the case of generating pressure only with a pneumatic pump by the comparative pressure generating experiment of pressure booster and only pneumatic pump. In a pressure tracking experiment using a simultaneous positive pressure and negative pressure controller, it was confirmed that the tracking performance was significantly improved when DOB was applied together as compared with the case where there was only a PID controller. It is expected to be improved as a technology applied to the power source system for pneumatic soft robots by combining pressure booster and simultaneous positive and negative pressure controllers.

Keywords: Soft Robotics, Force and Tactile Sensing, AI Reasoning Methods for Robotics
Abstract: This paper describes a sensing technology that transmits contact information to a robot in the event of a collision and a sensor that can physically absorb shock for safe collaboration between humans and robots. The large-area balloon sensor consists of a time of flight (ToF) sensor and an air layer cover layer, and includes an algorithm that recognizes changes in the cover layer through ToF sensor. A device for automatically pushing a cylindrical sensor iss manufactured. Using the data obtained through this as the correct answer data, supervised learning, one of the machine learning techniques, is used to match the ToF sensor data and position data of the pushing device when the object and the cover layer are in contact.

Keywords: Actuation and Actuators
Abstract: In recent years, nanohybrid organic-inorganic materials has drawn the scientific interest for variety of technological applications. One of the fine organo-inorganic materials with interesting dielectric, optical, flame retardant, and electrochemical properties is phosphazene-based materials. The core structural skeleton of phosphazene include discontinuous positioning of nitrogen and phosphorous atoms in the prime string. Phosphazene nanotubes (pnt) with functioning hydroxy moieties was prepared by templated synthesis under mild reaction route.The plasticizer used to polyvinylchloride (PVC) was acetyltributylcitrate (ATBC, ecofriendly plasticizer) and the content was optimized to PVC, the as-prepared pnt was stoichiometrically loaded in various quantities to fabricate soft and flexible phosphazene nanotubes comprising PVC nanocomposite (pnt-PVC) gels. The physicochemical characteristics and the interaction of pnt with plasticized PVC was subsequently investigated using advanced analytical techniques. The proposed pnt-PVC gels were utilized to construct the vibrotactile actuators as a dielectric interface layer and its actuation performances were evaluated.

Keywords: Actuation and Actuators, Soft Robotics
Abstract: Soft actuators can manipulate unknown objects of varying size and shape due to their flexibility, and especially pneumatic soft actuators are of particular interest for soft robotics because of their advantages. However, current pneumatic soft actuators have a limitation of the short-range of movement, which makes them difficult to move an object for a long distance. The present study proposes a new type of soft pneumatic actuator with a long-range of movement using the resonance phenomenon. To investigate the feasibility of the proposed actuator, we have done three experiments, and these are experimentally confirmed. 1) the occurrence of ellipsoidal trajectories. 2) the possibility of determining the direction of the ellipsoidal rotation. 3) the controllability of the direction of the surface traction.

Keywords: Actuation and Actuators, Soft Robotics, Human-Robot Augmentation
Abstract: Owing to the significant advantages on soft actuators, various researches on soft robots have been conducted actively. Especially, studies on prosthetic limb driven by soft actuators as artificial muscle are most representative. Also, Dielectric elastomer actuator (DEA) , shape memory alloy (SMA) actuator, and twisted and coiled actuator (TCA) are studied as artificial muscle. In this paper, active yarn actuator that can be used in fabric type actuator fabricated from SMA is proposed. Previously, SMA wire is processed in coil shape to improve strain performance. However, coil shape is not appropriate in fabric and wearable suit because they occupy large volume and weak on vertical stress. In this study, zigzag shape SMA (ZSMA) is developed to prevent the disadvantages of coil SMA and analyzed according to the parameters of ZSMA.

Keywords: Actuation and Actuators, Soft Robotics, Performance Evaluation and Optimization
Abstract: Recently, an exoskeleton has been developed to apply additional torque to the human joints or adjust the stiffness of the human body to assist human performance. They needs to be designed with two primary goals to achieve optimal form and function for safe and dynamic interaction. First, minimize the kinematic restrictions on the user's natural degrees of freedom. It is necessary to increase the user's safety with high backdrivability. Second, minimize the wearable robot weight to reduce the burden. An exoskeleton with quasi-passive actuation can be an attractive solution. It has been studied as simple structures using elastic springs and clutches. The spring stores elastic energy during the calf muscle relaxes and returns the stored energy with muscle contraction. The clutches control the energy storage timing to prevent interference of the elastic body during tension. This study developed a variable stiffness band that can change the stiffness according to the motion by embedding a clutch in the elastic body. Unlike existing quasi-passive actuation, which can provide a fixed auxiliary torque trajectory, it releases energy separately through a clutch at the desired timing and provides adjustable assistive torque. The electroadhesive clutch can provide a high brake force capable of actuating different stiffness with a small volume. Also, the developed variable stiffness band is customizable according to the target joint. Developed variable stiffness band performed 0.96 Nm/rad in dorsiflexion(DF) and 3.22 Nm/rad in plantarflexion(PF). In dynamic situation, VSB perform stiffness of 1.66 Nm/rad in the dorsiflexion, and 4.05 Nm/rad in the plantarflexion. The energy consumption rate in PF is about twice that of energy storage in DF.

Keywords: Aerial and Flying Robots, Performance Evaluation and Optimization
Abstract: Aerial manipulation systems have been an active research topic for recent years, but still have critical barriers such as flight time, inaccurate control/estimation, and insufficient payload to apply at manipulation tasks. In previous research, LASDRA (Large-size Aerial Skeleton with Distributed Rotor Actuation)cite{ICRA18LASDRA} was proposed to overcome these limitations. In this paper, we present experimental results of the 3-link LASDRA system with a joint locking mechanism to verify the performance and effectiveness of the system by performing trajectory tracking and 1.28kg object transportation experiments

Keywords: AI Reasoning Methods for Robotics, Object Recognition, Computer Vision and Visual Servoing
Abstract: Inspection of underground facilities is dangerous for a human being to work in dark environments and closed terrain. In this paper, we propose an integrated framework, which is an underlying technology, for mobile robots to inspect underground facilities. The proposed framework consists of equipment for collecting facility data in a tunnel and an object detection model that combines low-light enhancement, object detection, and tracking algorithms in a low-light environment. The proposed model can be used to diagnose underground facilities with mobile robots and shows remarkable object detection results even in a low-light environment.

Keywords: Computer Vision and Visual Servoing, Medical Robotics and Computer-Integrated Surgery
Abstract: Thanks to the accuracy and convenience, the modern surgical navigation plays an assisting role in operation rooms and performs necessary functions of surgical robots. The registration technology of surgical navigation is also critical to augmented reality. One of the essential technologies of augmented reality-based surgery is to accurately display 3D medical image of a patient on the patient’s location. This study proposes a markerless registration method that can be automated more conveniently than the conventional marker-based registration method. The proposed system, consisting of RGB-D camera and stereovision, allows for the simultaneous acquisition of the coordinates of both a patient and a surgical tool. The algorithm of the proposed system was validated by using the distance error between the 3D coordinates of a surgical target acquired from the stereovision and the patient’s coordinates transformed from the medical image coordinates and then registered.

Keywords: Force and Tactile Sensing
Abstract: In this paper, the glove which detects human hand motion of thumb, index, and middle fingers is proposed by using tendon-inspired piezoelectric sensors. The sensors were miniaturized to fit size of human hand and modified for attaching on the wrist brace. The sewing threads of 9 sensors were tied at middle of phalanx to detect rotation of the each joint. Also, other two sensors detected abduction/adduction motion between two fingers. Thus, 11 sensors were attached to the gloves. The sensors transferred voltage data to computer through the interface board. Moreover, the linking process between virtual reality hand and the glove was performed.

Keywords: Force and Tactile Sensing, Grasping, Mechanism and Design
Abstract: In modern society, there is a growing need to develop grippers that can precisely control works. For precise control, it is necessary to develop a gripper tooltip capable of sensing gripping force. To cope with this demand, research is underway on grippers and gripper tooltips that can sense gripping force. In this study, we propose a gripper tooltip with the same force sensing performance even if the type and shape of the work piece differ by utilizing a multi-support structure with flexible hinges.

Keywords: Force and Tactile Sensing, Robotic Hands, Grasping
Abstract: The tactile sensor based on barometric pressure sensor chip (BPSC) has been widely used as an inexpensive, sensitive, and easy-fabricable soft sensor for robotic gripper and hand applications. However, due to its mono-axial detectability, adaptive grasp and stable in-hand manipulation cannot be performed properly, especially when the object slip occurs. The aim of this study is to investigate the ability to detect slip with BPSC-based tactile sensor by processing the raw output data from the sensor. To observe the high frequency signal due to the vibration caused by the slip, discrete wavelet transform (DWT) is held in the order of 4. In addition, a threshold value is set to assess its capability. For a further understanding, the parametric analysis is experimentally conducted with three main parameters affecting slip detectability: the thickness of the cast soft material, velocity of the slippage, and amplitude of preload before initiating the slip. The results demonstrate the parametric effect on the slip-detection capability of BPSC-based tactile sensor.

Keywords: Grasping, Object Recognition, Manipulation Planning and Control
Abstract: In this paper, we propose a picking test objects based on the shapes and materials for random piece picking. The random piece picking solution is a core technology for automation of the logistic and manufacture. Evaluating the ability to pick or grip unseen objects is one of the challenge issues since the piece picking requires picking lots of different type objects. Therefore, this paper describes the selecton of test objects for the evaluation of picking point recognition performance with real robot. In order to select the test objects, we classified by the objects by shape and material and selected representative objects. Through this test objects, we can know whether the picking performance is affected by the material or shape of the object.

Keywords: Haptics, Contact: Modeling, Sensing and Control, Multisensor Data Fusion
Abstract: We propose a system of haptic rendering and interactive simulation which enables elaborate hand-object manipulation in virtual reality (VR). We unify the visual-inertial skeletal tracking (VIST)-based hand-tracking system, passive midpoint integrator (PMI)-based interactive simulation, and tactile feedback device, accomplishing a novel, stable and robust hand simulator in VR. The hand-tracking part employs VIST algorithm, tightly-coupled sensor fusion of vision and inertial sensors, which is robust against occlusion, interference, and contact. PMI-based simulation approximately preserves passivity and non-iteratively calculates each time-step, so that it enforces stability and real-time immediacy. We also verify this system implementing in two different scenarios: peg-in-hole and ball manipulation.

Keywords: Haptics, Human-Robot Augmentation
Abstract: This paper presents a new haptic system that can render virtual texture by providing multi-modal tactile feedback to a user. The proposed haptic interface has a set of vibrotactile actuators and a rotating textured tactile element, which can effectively mimic the tactile sensation of sweeping over a textured surface. Additionally, a force feedback interface provides additive kinesthetic feedback to a user's finger. We set up haptic rendering algorithms to evaluate the effect of each tactile cue on the haptic perception of virtual texture.

Keywords: Haptics, Modeling, Identification, Calibration, Medical Robotics and Computer-Integrated Surgery
Abstract: This paper shows an algorithm of distortion compensation for a cable-driven position device. A haptic device connected with four cables is fabricated for the operation of the surgical robot. By measuring the lengths of the four cables connected to the grip, the position of the center of the grip in space can be estimated. In this case, distortion such as a barrel image in the image field occurs when estimating a position in a direction perpendicular to the plane in which the four string pods are located. After the distortion phenomenon is projected onto the ellipsoid and modeled, the position in the vertical direction of the cable plane is compensated by the corresponding value and flattened.

Keywords: Industrial Robots, Dynamics and Control
Abstract: In this article, we propose a temperature compensation model for a robot manipulators which can be applied to an outdoor installed robot in autonomous electric vehicle(EV) charging system. The traditional model-based collision detection method does not guarantee a consistent performance if the robot is installed outdoor due to a different environmental conditions such as outdoor temperature. It is obvious that the inner temperature of joint actuators is affected by the outer temperature and operating time of robot which directly affect the dynamic model and frictional torque. Therefore it is necessary to deal with the temperature issue to use the robot in outdoor EV charging system without loosing model precision. We estimated the dynamic parameters of the robot at different temperature conditions using a torque separation technique and regression matrix. Then our proposed temperature model for friction parameters were validated via experiments. The experimental results shows effectiveness of temperature compensation method.

Keywords: Industrial Robots, Learning From Humans, Manipulation Planning and Control
Abstract: Currently, the performance of commonly used cooperative robots through learning form demonstration(LfD) is mainly performed such as pick and place based on point to point movement. However, in reality, there are much more diverse and complex tasks than that, and in order to perform them through robots, other movements and control techniques for them must be met. The reason why it is difficult to perform the task of contacting the object is that two control methods must be satisfied: force control and following the task trajectories. To perform two conflicting control methods at the same time, a continuous trajectory was stably followed through the Dynamic Movement Primitives(DMPs) algorithm, and in the process, the target force was followed through adaptive impedance control. Two practical experiments requiring trajectory and target force tracking were conducted to evaluate how well the target trajectory and force are followed. And based on the results in the experiment, we examined what remarkable results and what limitations were revealed.

Keywords: Industrial Robots, Motion Planning and Obstacle Avoidance, Manipulation Planning and Control
Abstract: Robot welding is an existing application in automobile manufacturing and shipbuilding. Scanners are actively used to guide industrial robots to precisely follow target welding lines. Until now, most robotic welding has been automated for small-scale welding (eg automotive parts or frames) because the robot used for welding is stationary. It has been considered difficult to use a mobile manipulator, which has been actively developed in recent years, for high-precision welding. The working precision of the mobile manipulator is significantly lower than that of the fixed type, and the precision of the movement itself is poor and slow. For this reason, it is difficult to use in industrial fields. Nevertheless, the welding operation time for large objects such as the LNG cargo containment, which is the subject of this study, is incomparably long (several months), and since it is a tough operation for humans, there is a great advantage that can be achieved through welding automation using robots. Here we aim to develop offline programming software for mobile manipulators that runs in a huge space with a floor. Welding precision and robot work area precision are achieved by different means in a narrow work area. The purpose of this study is to create a full robot motion that tracks a complex welding vessel in a multi-layered LNG tank.

Keywords: Intelligent Robotic Vehicles, Multi-Robot Systems, Motion Planning and Obstacle Avoidance
Abstract: This paper addresses the development and preliminary tests of multiple autonomous surface vehicles (ASVs) for autonomous cooperative navigation tasks in an inland water. The developed ASVs are designed as a catamaran hull form, and various systems related with autonomous functionalities, including the electrical propulsion system, the wireless communication system, the guidance, navigation and control (GNC) system, are integrated with perception sensors. In particular, for cooperative navigation between ASVs, the network communication system is implemented to be shared with its each motion information through ROS multi-master system. An autonomy framework with a cooperative navigation and control is designed to follow the predefined waypoints as well as to keep a specific formation based on the distance between multiple ASVs. To demonstrate a essential maneuvering and functional capabilities for cooperative navigation of the developed ASVs, preliminary field tests were carried out in the inland water environment, and its results were described.

Keywords: Intelligent Robotic Vehicles, Multi-Robot Systems, Range, Sonar, GPS and Inertial Sensing
Abstract: Autonomous surface vehicles (ASV) are widely used to perform various tasks in marine environments. Especially, multiple ASVs can cooperate to accomplish given tasks efficiently. This paper presents a method of cooperative positioning system for multiple ASVs. When an ASV is in trouble with GNSS based positioning system, another ASV with accurate position can be used as a landmark to localize the ASV. A localization method for each ASV is basically implemented by INS-GPS fusion. Furthermore, each ASV estimates absolute position of other ASV. From this concept of cooperative positioning system, the ASV which experiences unavailable case of GNSS data estimate its own location using relative geometric information to other ASV. The proposed method is implemented by Kalman filter based estimation and verified by inland field tests.

Keywords: Manipulation Planning and Control, Industrial Robots, Dynamics and Control
Abstract: Compliance control is essential for tasks which requires a robot manipulator to interact with the environment, such as deburring, grinding or human-robot collaboration. Admittance control is a widely used compliance control methods, especially for position-controlled industrial robots. However, one of the drawbacks in implementing admittance control in real practice is the contact stability. In this study, model predictive variable admittance control is proposed to cope with the contact stability of robot manipulators. Using variable admittance control, the virtual damping and stiffness of an admittance controller can be changed online. Model predictive control (MPC) is applied to optimize the virtual damping and stiffness, such that the vibration upon contact can be minimized. The feasibility of the proposed controller is verified through simulations with a 3 DOF robot manipulator.

Keywords: Manipulation Planning and Control, Motion Planning and Obstacle Avoidance, Industrial Robots
Abstract: Recently, manufacturing automation scope is rapidly expanding for rigid object, not for soft object like cable wiring. This is because that handling flexible cable and detecting it is difficult. Thus, a method is needed to decrease the dependence of the visual sensor. So, this paper introduces a strategic algorithm that can minimize dependency about image sensor, and friction effect for the algorithm. A friction model was defined considering form between the pin and the cable generated during the algorithm implementation, this paper was figured out a correlation between winding angle and direction for successive cable wiring.

Keywords: Mechanism and Design
Abstract: One-degree-of-freedom (DOF) underactuated spring-linkage mechanisms can be adaptive in a 2-D plane if their inside spring elements properly deform. We propose a method to synthesize 1-DOF underactuated spring-linkage mechanisms to follow a given target path even when there is a possible obstacle along the route. We find the link dimension and the layout distribution of spring elements in a rigid-link mechanism of a pre-selected topology through the method. To obtain a quite arbitrary spring layout, we use a spring ground model consisting of many spring elements connecting links of the baseline mechanism. 1-DOF underactuated spring-linkage mechanisms generating adaptive gait motions are considered as synthesis examples.

Keywords: Mechanism and Design
Abstract: We design the 2 degrees-of-freedom(DOF) robotic wrist with an aim of developing an active flat-type compliant joint capable of being mounted between the robotic arm and end-effector and so improving the manipulability. To achieve this aim, the sheet-type diaphragm flexure mechanism is applied as a motion guide of the wrist. When the robotic end-effector performs various grasping/manipulation tasks accompanying physical interaction with the environment, it is necessary to reduce the physical damage caused by undesired collision. The wrist mechanism connecting the arm with the end-effector structurally can provide the function to attenuate the external shock transmitting to the manipulator. However, it is not suitable in general to apply an existing robotic wrist for this function, due to its heavy weight and large size in the longitudinal direction. Considering this limitation, we attempt to realize the lightweight flat-type robotic wrist having 2-DOF tilt motions. As a key component allowing the wrist to have these characteristics, we design the diaphragm flexure. In addition, the parametric effect of main design parameters of the mechanism on the operational specifications such as a range of motion and effective stiffness is investigated based on finite element simulation.

Keywords: Mechanism and Design, Actuation and Actuators, Social and Socially Assistive Robotics
Abstract: With the decrease in the number of agricultural workers and the rapid aging of the population in Republic of Korea, the need for robot technology to assist human work to minimize the risk of disease and enable efficient work has been emerged in the agricultural field. In various fields, robots and devices for disease prevention and treatment of workers or patients have been developed by applying wearable robots. However, there is a limit to their application to agricultural workspace in which unstructured work is mixed. In addition, the conventional technologies apply a control that requires various sensors to understand the user's intention, so that there are problems such as high-spec system for data processing, short battery life, high cost, heavy weight product, and difficulty in wearing. In order to solve these problems, it is essential to develop an innovative driving mechanism. In this study, we explain the simultaneous bidirectional driving mechanism and the design of the bidirectional actuator module for various robot systems, and verify the mechanism and performance of the proposed simultaneous bidirectional actuator module by experiments.

Keywords: Mechanism and Design, Grasping
Abstract: This paper introduces a new grasper for the automation of a logistics palletizing process. This grasper operates by a flexible and thin band making a vertical loop around a target object. The grasper consists of two separated parts, an upper band driver which is a part of the palletizing robot and a lower band guider which is fixed to the conveyor system. The band extruded from the band driver passes through the band guider, and makes the vertical loop around the target object. Because the proposed grasper supports the bottom and side of the object, it can have a large payload and be robust to horizontal accelerations. In addition, the grasper can place the objects effectively in a narrow space because a thin band is used. The capability of the grasper was verified through an experiment with six objects. Corrugated box, plastic box, packaged water, plastic drum, square aluminum drum, and EPS box were all successfully gripped as intended.

Keywords: Mechanism and Design, Medical Robotics and Computer-Integrated Surgery, Rehabilitation and Healthcare Robotics
Abstract: Above-the-knee amputees expend significantly more metabolic energy for walking than non-amputees. In order to reduce the energy consumption of the amputees, a robotic knee prosthesis is required. The robotic knee prosthesis helps natural gait and greatly reduces the amputees' energy consumption but may cause impact to the user when the foot is placed on the ground if it has no elasticity. Therefore, compliance modules are required for energy absorption and power reduction. In this paper, we propose a robotic knee prosthesis with a serial elastic actuator. The gait cycle is divided into stance phase and swing phase, and the robot prosthesis is controlled by applying impedance control and position control in each phase. The compliance module is used in the stance phase.

Keywords: Mechanism and Design, Physical and Cognitive Human-Robot Interaction, Industrial Robots
Abstract: Various types of counterbalance mechanisms have been suggested to enable humans and robots to coexist safely. In this study, an advanced adjustable counterbalance mechanism with multiple degrees of freedom (DOFs) using a gear roller and springs was designed. The reliability was improved compared to the previous wire-based counterbalance mechanism. Moreover, the torque required to support the self-weight of the robot arm and any additional external load was compensated for using the proposed adjustable counterbalance mechanism. To verify the effectiveness of the mechanism, an unpowered crane based on a 2-DOF adjustable counterbalance mechanism was constructed and demonstrated.

Keywords: Mechanism and Design, Search and Rescue Robotics, Modular Robots
Abstract: Supplies of food, water, and medicine can significantly enhance the chance of survival for a person trapped in narrow collapsed structures. This paper proposes a compact liquid supply system for a snake-like robot to rescue trapped survivors in collapsed structures. The proposed liquid supply system consists of a liquid supply module in the tail part of the snake robot, three tube-lines that can transport liquid from the tail to the head of the snake robot, and nozzles located in the head module of the snake robot. The liquid supply module can provide three liquids: water, juice, and liquid medicine. The effectiveness of the proposed liquid supply system for a snake-like rescue robot is verified experimentally.

Keywords: Mechanism and Design, Soft Robotics
Abstract: In order for wearable robots to be used socially, the concept of embedding an exoskeleton in normal clothing is suggested. Therefore, in previous work, we presented the concept of smart clothes in the form of daily clothing by attaching the actuator to normal clothes. In the previous paper’s smart clothes, the actuator connects the ankle from the waist. In the meantime, the actuator passes the knee which joint shows a large range of movement from 0° to 90°. In order to minimize the inconvenience of smart clothes and increase the efficiency of the actuator during the operation, experiments were conducted to design actuation path routing considering human ergonomics and force transmission.

Keywords: Mechanism and Design, Soft Robotics, Modeling, Identification, Calibration
Abstract: This paper is about optimizing the elbow assist torque of a Soft Wearable Robot (SWR) developed for the purpose of reducing musculoskeletal disorders in industrial workers. The developed SWR will supply the elbow auxiliary torque to the person from the motor and tendon-wire. At this time, the hinge point and the point of application were determined so that the elbow assist torque was maximized by modeling the movement of the human elbow and wire routing. The mathematically modeled human upper body and SWR were decorated with an experimental device and the reliability was verified by comparing it with the actual motor torque. The parameters and modeling used in this study were used for the actual SWB production, and in the next paper, we will verify the muscle mass fatigue of a person wearing the SWB by measuring the EMG sensor signal.

Keywords: Mechanism and Design, Underwater Robotics, Robotics in Hazardous Applications
Abstract: The substructure of an offshore wind turbine requires periodical maintenance and inspection, but working conditions are harsh due to its large and complex structure located under water. Thus, first, in order to have fast mobility, a wheel-type pole climbing robot was chosen. In addition, the slip problem, which is a disadvantage of the wheel type platform, was supplemented by applying a gripper and self-locking system to the design. Using these concepts, a new dual type platform J-UPCR (Jeju national university-Underwater Pole-Climbing Robot) was developed to climb a branched pole structure. As a result of experiment, J-UPCR was able to climb the branched poles and maintained its position without consuming energy. Therefore, it was verified that the dual platform, J-UPCR was turn to have the basic mobility performance necessary for working in the complex substructure of an offshore wind turbine, which is located in underwater environment.

Keywords: Mechanism and Design, Wheeled Mobile Robots
Abstract: With the growth of the e-commerce market, the demand for last-mile delivery robots is increasing. Most of the last mile delivery robots use four-wheels, which is difficult to overcome stairs. There is a problem in performing delivery. The mechanism of the Eccentric Crank Rover(ECR) demonstrate excellent performance in overcoming stairs. But, when a load is added, flat ground driving energy efficiency decreases. This paper introduce WAVE, Last-mile delivery robot using Four-bar linkage mechanism with sawtooth structure designed for the shape of the stairs. In order to increase the flat ground driving energy efficiency, In-wheel is added, which can change the driving mode. In addition, this paper includes stair-climbing experiment.

Keywords: Mechanism and Design, Wheeled Mobile Robots
Abstract: In this paper, we propose a mobile robot structure that has as high maneuverability as AWS(All Wheel Steering) does but has a simpler mechanical structure. Normally, 4-wheel AWS robots have 4 driving motors and 4 steering motors for each driving motor. We propose a structure that uses just a single motor and a linear actuator to steer driving motors. The proposed structure can do counter-steering, rotate in-place, and move horizontally. In other words, it has enough maneuverability for tasks required in a smart farm. Also, it has fewer motors than AWS, so we can manufacture the system simpler and cheaper.

Keywords: Medical Robotics and Computer-Integrated Surgery, Computer Vision and Visual Servoing, Object Recognition
Abstract: This paper presents an algorithm for automating swab sampling of upper respiratory diseases. In particular, by applying deep learning-based object detection technology, the image data that the target objects overlapped at a certain location was learned as a specific "state", and the state was detected by the detection algorithm. The detected state is then used for automatic swab positioning by a robot. This paper introduces the system configuration to implement the algorithm, a flowchart applying the proposed method to automate swab positioning in front of the nostril, and an automatic swab sampling experiments using the phantom model were introduced.

Keywords: Medical Robotics and Computer-Integrated Surgery, Mechanism and Design
Abstract: A core biopsy is a key procedure in diagnosing clinically significant prostate cancers. Real-time magnetic resonance imaging (MRI)-guided biopsy offers various advantages over standard transrectal ultrasound (TRUS)-guided biopsy, but it carries tremendous challenges as well. Automation of MRI-guided biopsy requires teleoperated MRI-safe robots with the ability to take and store the core samples. A device that can take tissue samples from the needle cavity and store temporarily has been proposed and is under development in our research group. The proposed multiple tissue sample collection device (MTSCD) uses a pair of moving paper sheets for collection as well as storage. This work was done to see the effect of the tension in the paper on the performance of the device. Various tests were performed for different tension values and results were compared using an image processing technique. Optimization of further variables is ongoing to improve the performance of the device.

Keywords: Medical Robotics and Computer-Integrated Surgery, Mechanism and Design, Rehabilitation and Healthcare Robotics
Abstract: In this paper, a new type of robotic system for non-invasive brain stimulation (NIBS) is proposed for the patient safety with high accuracy. We applied the Stewart platform and remote center of motion (RCM) mechanisms with the NIBS robot. Six degrees of freedom (DOFs) of the Stewart platform enables the brain stimulator to be positioned more accurately and quickly compared with the previous serial robots. To compensate a relatively small workspace of the Stewart platform compared to the required workspace for NIBS, two DOFs of the RCM mechanism were applied. Since the proposed RCM mechanism is designed to limits the rotational ranges, it reduces the risk of collision between a patient and robot. Furthermore, the revolute joint of the RCM mechanism has high torque efficiency compared with the conventional mechanism. Through the experiments, we confirmed that the proposed robot reduced the position tracking error by approximately 20% and the required motor torque at the revolute joint by 158 times compared with the conventional robot. The proposed NIBS robot is the first successful trial based on the Stewart platform and RCM mechanism that ensures patient safety, precision, and torque efficiency.

Keywords: Medical Robotics and Computer-Integrated Surgery, Robotic Systems Architectures and Programming
Abstract: This paper proposes a non-contact otolaryngology diagnostic robot system based on UX design. Since the outbreak of Covid-19, many non-face-to-face robots and diagnostic systems have been developed for triage room. The proposed system can be applied not only at triage room, but also for general otolaryngology or respiratory disease diagnosis, and conventional hospital system as well. Thus, UX/UI design issue is essential to localize the pain points of both patients and doctors and to provide an efficient and comfortable experience for both doctors and patients. In light of such points, this paper describes the overall architecture and service scenario of Otolaryngology diagnostic Robot System.

Keywords: Micro/Nano Robots, Soft Robotics, Computer Vision and Visual Servoing
Abstract: Imaging flow cytometry is an essential biomedical research tool that combines the single-cell imaging capabilities of microscopy with the features of conventional flow cytometry. Recent advances in deep learning and high-throughput optical imaging enable deep learning-based imaging flow cytometry. However, the long data processing time of deep learning and image processing pipeline is a bottleneck to develop a high-throughput imaging flow cytometry. The present study examined a high-speed imaging system for detecting and classifying particles in a fluid flow under the TensorRT deep learning framework. To investigate the feasibility of the proposed system, the processing time was measured and a real-time classification of polystyrene beads was carried out. The proposed system can be combined with a sorting device, such as a piezoelectric actuator, to perform real-time cell sorting or cell enrichment.

Keywords: Multi-Robot Systems
Abstract: This paper presents the development of multiple unmanned surface vehicle system(USV), which consists of three USVs. Each USV is equipped with a lidar, a radar, and two cameras and is designed to have a capability to detect surrounding objects around it. In such multi-robot systems, handling network loads is an important design consideration; each vehicle is required to share its on-board sensor data with each other, but not all of data can be shared due to network load. Therefore, it is important to design the hardware and software architecture considering which data is shared inside each USV, and which data is shared between the vehicles. In this paper, the network structure design of the multiple USV system, which adopts ROS multimaster is introduces and the result of its preliminary field experiments are presented.

Keywords: Multi-Robot Systems, Industrial Robots
Abstract: This paper proposes a system framework that can be applied to multi-agent-based real-world robotic manipulation systems in the field of manufacturing. The proposed system framework enables real-world robotic manipulators to perform a cooperative task not only through calculated trajectories by an inverse kinematics solver in real-world environments but also through learned behaviors by deep reinforcement learning algorithms in simulated environments. Here, the cooperative task refers to a task that requires two or more robotic manipulators to perform the task. To this end, the system consists of four main blocks: multi-agent-based real-world robotic manipulation ROS framework, real-world robotic environment, simulated robotic environment, and sim-to-real interfacing. The system has been tested in real-world environments under two cooperative tasks ``peg-in-hole'' and ``pick-and-stack'' to validate its feasibility on multi-agent-based real-world robotic manipulation.

Keywords: Neurorobotics, Physical and Cognitive Human-Robot Interaction, Telerobotics
Abstract: When a human performs a given specific task, it has been known that the central nervous system controls modularized muscle group, which is called muscle synergy. For human-robot interface design problem, therefore, the muscle synergy can be utilized to reduce the dimensionality of control signal as well as the complexity of classifying human posture and motion. In this paper, we propose an approach to design a human-robot interface which enables a human operator to transfer a kinodynamic control command to robotic systems. A key feature of the proposed approach is that the muscle synergy and corresponding activation curve are employed to calculate a force generated by a tool at the robot end effector. A test bed for experiments consisted of two armband type surface electromyography sensors, an RGB-d camera, and a Kinova Gen2 robotic manipulator to verify the proposed approach. The result showed that both force and velocity commands could be successfully transferred to the robotic manipulator via our muscle synergy-based kinodynamic interface.

Keywords: Object Recognition
Abstract: Deep learning models with a small number of parameters (small model) have an important role in edge com- puting devices such as mobile robots. However, the performance of a small model is not guaranteed compared to the large model. In this paper, we present a training method called attention mask guided learning, which is for improving the performance of small deep learning model on edge computing devices. In our proposed method, we supposed that a large model which is pretrained by big data on a cloud system provides an attention mask to guide the training of a small model. The attention mask generated from the region of interest of a large model focuses on the target object by ignoring the background. Then, the raw image and the masked image guided by attention mask are provided into the small model, obtaining the features of each input. Using features, we add feature comparison as an auxiliary loss to ignore the background of the input image. Through feature comparison, the small model ignores the background by the result of the raw image and the image ignoring the background are induced to be similar, so that even if the raw image is entered, the background is ignored and the target object is focused. As a result, the performance of MobileNetV3 is improved by 2.8%, 0.7%, and 2.0% in CUB, Stanford Cars, and FGVC-Aircraft dataset respectively, compared to the from-scratched model. The experimental result shows that our method leads the small to be trained by focusing on the target object in the image, so the classification performance is improved compared to when the method is not applied.

Keywords: Object Recognition, Computer Vision and Visual Servoing
Abstract: Out-of-distribution (OOD) detection is considered one of the important problems to solve for the deployment of deep neural networks. In this paper, we propose the method for OOD detection: the training with a normalized feature is based on the idea that large norm of the feature causes the overconfident prediction even for a given OOD image; the scoring method with a norm of the feature is based on the previous observation that the norm of the feature represents the quality of the input. Our method is simple yet effective, achieved comparable results on CIFAR benchmarks. Our analysis of our method showed that the norm of the feature for given OOD samples is almost zero, while the norm of the feature of ID is much larger. Since it is a very simple approach, we believe this method can be further studied for many applications.

Keywords: Object Recognition, Computer Vision and Visual Servoing
Abstract: Out-of-distribution (OOD) detection, which aims to distinguish whether the test image is an image that belongs to one of the trained categories (i.e., in-distribution) or not (i.e.,out-of-distribution), is an essential task for deploying the deep neural network in real-life. We argue that auxiliary models which are off-the-shelf models that are trained on another dataset which from similar domain can be helpful for OOD detection of the main model. Our insight is that features of ID for the main model will be close together even in the feature space of auxiliary models, while OOD images which from different domain are on the other area. Therefore, our proposed method to detect OOD of the given image is a sum of the Euclidean distance between the feature of the test image and the pre-calculated class-center. Our experimental results on the CIFAR100 benchmark demonstrated that our method can outperform previous methods. To the best of our knowledge, this is the first attempt to utilize the auxiliary models in OOD detection research, and we showed that auxiliary models are helpful for OOD detection. Our future work will focus on weighting methods in auxiliary models with the degree of assistance.

Keywords: Object Recognition, Multi-Robot Systems, AI Reasoning Methods for Robotics
Abstract: The robotics industry is largely divided into two categories, either replacing work done by humans or enabling challenging work to accomplish tasks that humans cannot. Among them, in the robot industry that replaces human work, it is important how well human actions or tasks can be implemented and can be done accordingly. Logistics centers need to reduce costs, improve efficiency, solve the problem of manpower shortage (at night/peak time/season of rush orders), and improve job quality through automation and intelligence. Here, in order for the robot to manipulate it as human order picking, dataset for Deep Learning Network is essential for supervised learning. Therefore, for this, this work make a Dataset Concept for Learning Deep learning based Picking Algorithm.

Keywords: Physical and Cognitive Human-Robot Interaction, Manipulation Planning and Control, Force and Tactile Sensing
Abstract: To realize efficient human-robot-interaction (HRI), an intuitive communication method between humans and robots is required. In this paper, new intuitive human interface device for robotic applications is introduced for realizing the HRI. The sensing principle and design of the device are mainly explained. The device can detect a six-axis force/torque (F/T) with a six degree-of-freedom displacement. The device mainly consists of a handle, a sensing part, and a base frame. The sensing part was designed with capacitance sensing principle using an eccentric arrangement of electrodes. For ease of use, the part was designed to mechanically limit deformation by applying a blocking mechanism. The designed device has all components built-in and is manufactured in the form of a joystick that can be easily operated by a human.

Keywords: Physical and Cognitive Human-Robot Interaction, Social and Socially Assistive Robotics, Mechanism and Design
Abstract: Social robots must constantly interact with humans to perform physical or intelligent tasks. Research on how to design robot movements is important for efficient human-robot interaction. This paper applies the upper body movement among basic human movements to social robots. To find out whether the robot's motion is appropriate, we performed a user preference verification evaluation on the social robot's motion information design. Then, we modified the design for the motion information with a low preference score and finally derived the motion HRI of the social robot.

Keywords: Physical and Cognitive Human-Robot Interaction, Social and Socially Assistive Robotics, Performance Evaluation and Optimization
Abstract: Recently, museums are diversifying the viewing method beyond the existing passive knowledge transfer viewing method to provide the visitors with active information, and the trend is changing to the edutainment aspect to induce the visitor's participation. In this paper, the conversational agent interface technology for providing a variety of information actively uses the robot system in the Human Robot Interaction (HRI) and the augmented reality (AR) system in the Human Computer Interaction (HCI) aspect and we analyzed how it affects the visitors. The user evaluation of the interaction between the museum and the visitors confirmed the superiority of the two viewing methods that provided various audio-visual interactions compared to the conventional one-way viewing method. The use of robots and AR system in the museum in terms of edutainment confirmed excellence in sustainability, process satisfaction, emotion satisfaction, and aesthetics, except for professionalism and responsiveness.

Keywords: Physical and Cognitive Human-Robot Interaction, Social and Socially Assistive Robotics, Performance Evaluation and Optimization
Abstract: It is important for social robots to build emotional bonds with people through interactions. In this study, style comments generated on the user's attire were applied to the social robot service, the performance was verified in real environment for the elderly, and user satisfaction with the service was examined.

Keywords: Rehabilitation and Healthcare Robotics, Grasping
Abstract: People who suffer from lost hand mobility such as stroke, spinal cord injury, cerebral palsy, or other neuromuscular diseases, have different hand sizes and joint characteristics. Due to the distinctive finger joint characteristics, the moving or grasping performances of wearable hand robots vary according to the hand status of the wearer. To overcome this issue, it is important to develop a device that can measure and analyze the finger joint characteristics. To understand the finger joint characteristics, various devices were developed in previous studies, but have bulky designs and are limited to a specific hand with a specific injury. In this paper, the prototype of a finger joint stiffness measuring device that can measure metacarpophalangeal (MCP), proximal interphalangeal (PIP), and distal interphalangeal (DIP) joint characteristics of an index finger is presented. The experiment was performed on a healthy subject with five repetitions to show the consistent measuring capability of the device.

Keywords: Rehabilitation and Healthcare Robotics, Physical and Cognitive Human-Robot Interaction, Legged Robots
Abstract: Powered exoskeletons promise to improve mobility, but the research is progressing slowly, in part, as it requires extensive experiments with human subjects. We aim to develop a neuromechanical simulation framework that can predict the effect of leg exoskeletons on gait performance. Here, we lay out our plan and preliminary results.

Keywords: Rehabilitation and Healthcare Robotics, Social and Socially Assistive Robotics
Abstract: Motion Capture System (MCS) is used to track three-dimensional gait trajectories, and this study suggests a wearable device that can replace it using Ultra-Wideband (UWB). The device is equipped with a total of eight UWB sensors, four on each shoe, and calculates the relative position between each sensor using a trilateration method to create a three-dimensional gait trajectory of each foot. The foot trajectory estimated from our method using UWB sensors was very close to that measured with a motion capture system with a mean absolute error of 21.67 mm.

Keywords: Robot Surveillance and Security, Mechanism and Design
Abstract: As robot intelligence improves, the demand for various types of security robots is increasing. In particular, there is a growing demand for a platform that can freely move indoors and outdoors and perform security missions. Therefore, in this paper, we propose a mobile robot platform design method that can be operated in a wide range of indoor and outdoor environments. Part of the proposed system was verified through experiments.

Keywords: Robotics in Hazardous Applications, Actuation and Actuators, Mechanism and Design
Abstract: This paper proposes a hydraulic rotary servo valve and miniaturized version to the hydraulic robot. Unlike the existing flow servo valve, the electric motor is directly connected to drive the spool position, controlling the output flow rate. Hence, the proposed servo valve can solve a complex dynamic equation without an electro-hydraulic amplifier (Nozzle flapper or hydraulic jet pipe, etc.). Moreover, due to the lack of quiescent fluid flow in the electro-hydraulic amplifier, temperature-induced model uncertainties can be solved, and performance evaluation can be more accurately using the spool position. The feasibility test of first and miniaturized second version will be introduced in this study. Also, in nearly future, the control performance of second version will be evaluated.

Keywords: Robotics in Hazardous Applications, Multisensor Data Fusion
Abstract: Recently, research on operating robots for broadband resource exploration and reconnaissance in extreme cold environments such as Antarctica is attracting attention. Robots for autonomous driving and mission performance in extreme cold areas require sensor data acquisition and temperature maintenance methods that can operate in cryogenic environments. In this paper, we propose a method for acquiring multi-modal sensor data in a cryogenic environment and maintaining the temperature inside the robot. The proposed method tested for a long time in a cryogenic freezer to show its utility.

Keywords: Simultaneous Localization and Mapping (SLAM)
Abstract: The recent development in computer vision, odometry plays an increasingly important role in the field of autonomous systems. However, the traditional visual odometry for single camera or simultaneous localization and mapping only performs better in simple environments with obvious structural features. In a complex environment, visual odometry may fail due to the sparsity of stable features, sensor failure, and extreme weather conditions. Adding more camera information could easily solve such issues. However, less overlapped region between camera cause scale issue. Therefore, the multicamera LiDAR fusion algorithm is proposed. The features are extracted from each of camera image and the depth is registered from LiDAR for each camera image. The perspective-n point algorithm is used to estimate the camera motion. The pose and 3D points are optimized based on local bundle adjustment. The evaluation is performed based on GNSS provided ground truth, the proposed method showed significant improvement in challenging conditions.

Keywords: Simultaneous Localization and Mapping (SLAM), Intelligent Robotic Vehicles, Computer Vision and Visual Servoing
Abstract: In simultaneous localization and mapping (SLAM), the detection of true loop closure benefits in relocalization and increased map accuracy. However, its performance is largely affected by variation in light conditions, viewpoints, seasons, and presence of dynamic objects. Focusing the advantages of visual semantics to achieve human-like scene understanding, this research investigates semantics aided visual loop closure detection methods and presents a novel hierarchical loop closure detection framework for visual SLAM systems. Semantics being invariant to viewpoint changes and dynamic environment can improve the overall performance of the environment.

Keywords: Soft Robotics
Abstract: The robotic gripper including impactive-type and astrictive-type has been an essential component in the robot to perform certain tasks. The gripping capability of gripper mainly influences to the establishing efficient manufacturing process. Especially, the suction gripper which is belong to the astrictive-type gripper is one of the widely applied gripper in the industrial field due to its simple structure. However, these previous suction gripper has limitation to gripping object which has through hole or complex surface geometry. In this study, we proposed the honeycomb-shaped suction gripper which is capable of shape adaptation to the complex shape of target object.

Keywords: Soft Robotics, Actuation and Actuators, Grasping
Abstract: In this paper, we propose a new soft gripper using a hydraulically amplified self-healing electrostatics (HASEL) actuator in a stacked structure to grip a variety of objects. This actuator is manufactured in a small, basic rectangular shape, and uses hydraulic force and volume expansion generated when high voltage is applied. An amplified angular amount and force were generated by stacking one actuator in multiple stages. And it returns to its original state when it is turned off. The two types of soft grippers using the stacked HASEL actuator are tested and we expect that it can be used in various fields such as smart farm systems that require usability and mobility.

Keywords: Soft Robotics, Biomimetic and Bioinspired Robots
Abstract: There are significant demands to develop of soft crawling robots that can adapt to various unpredictable environments. To facilitate the development of soft crawling, new actuator modules based on soft materials and structures that can increase adaptability to terrain are needed. In this study, we propose new soft electrohydraulic actuator modules. The crawling robot which consists of front and rear legs, and two modules, operates similarly to the biomimetic mechanism of a caterpillar and implements forward motion. This soft crawling robot provides new design ideas for mobile exploration, manipulators, and biomimetic systems.

Keywords: Soft Robotics, Biomimetic and Bioinspired Robots, Wheeled Mobile Robots
Abstract: In this paper, we designed a bio-inspired, wheeled soft robot for adaptation capabilities of unstructured environments. This robot is composed of a cylindrical pneumatic body and four rigid ratchet wheels. The cylindrical body consists of left and right air chambers. The robot body can generate a gecko-like deformation by alternately pressurizing and depressurizing the two air chambers. The wheel was introduced to the design of the robot to improve its mobility and achieve barrier-crossing capabilities. We demonstrate that the robot can overcome obstacles and crawl on different terrains with benefits from both the gecko-inspired locomotion and the ratchet wheels.

Keywords: Soft Robotics, Grasping, Actuation and Actuators
Abstract: This work introduces the concept, design, and manufacturing of the suction cup inspired from octopus. This suction cup can grasp objects using the change in volume of membrane. This suction cup has capability to transport the pressure upto -77 kPa in air. Using syringe, the suction cup can achieve -51 kPa. This suction cup has potential to be used as an array to maximize its adaptability

Keywords: Soft Robotics, Haptics, Force and Tactile Sensing
Abstract: This paper proposes a design of self-sensing electrohydraulic actuator powered by induced interfacial charges (EPIC) via capacitance sensing circuit. EPIC was designed to be a powerful soft actuator with high performance for output force and low input voltage. We improved it to a smart actuator by adding self-sensing feature to extend its application as a smart haptic device that allows interactive operations based on external haptic stimuli and feedback-controlled varifocal lens. The self-sensing was accomplished by sensing the capacitance variation of the actuator, where a mixed-signal-based self-sensing method is adopted. Prior to designing the self-sensing circuit, electrochemical impedance spectroscopy was conducted to analyze the electrical characteristics of EPIC. From the result, the equivalent circuit of EPIC under DC-biased operation condition was modeled as a parallelly connected resistor and capacitor. In addition, the relationship between the contact electrode area and capacitance of the EPIC actuator was found to be proportional. Based on the electrical characteristics of EPIC, we designed a sensing circuit to measure contact force in accordance with the capacitance variation of the actuator.

Keywords: Soft Robotics, Mechanism and Design, Actuation and Actuators
Abstract: This work presents the concept, design and fabrication of a thermal pneumatic actuator driven by air thermal expansion. This actuator consists of an internal pneumatic artificial muscle (PAM) and an external heating chamber. The internal PAM is contracted and driven while the air pressure of the external chamber is increased by heating. This actuating is designed to make a linear motion. Based on the ideal gas equation, the increased pressure and force could be calculated by measured temperature. Using this concept, the pneumatic actuator could be driven without an external pneumatic pump, which can remove the noise and space restrictions.

Keywords: Soft Robotics, Mechanism and Design, Actuation and Actuators
Abstract: Shape memory alloy (SMA) actuators have been widely used for robots owing to their advantages of flexibility and light weight. However, there is a drawback that the SMAs need to be used as a bundle due to their lower force density than the motor. In addition, the SMAs require continuous current to maintain contracted state, which results in the overheating of the actuator causing breakdown and excessive energy consumption. This study presents new locking system which can hold the contracted state of the actuator mechanically without consuming the energy to overcome these disadvantages driven by a bundle of spring-type SMA actuator. The locking mechanism is designed with a ratchet to prevent the actuator from stretching in the opposite direction when locked and to switch locking and releasing with a single operation through a retractable mechanism. The finger applied to proposed locking system is fabricated and its feasibility is experimentally validated.

Keywords: Soft Robotics, Mechanism and Design, Dynamics and Control
Abstract: As society advances, robotic systems are becoming an integral part of society and the human experience. These include but are not limited to, articulating constructs for manufacturing, systems that assist with surgery or rehabilitation, or animatronic systems that entertain people. These systems and the others not mentioned, provide several beneficial services to society. However, the most important service robots provide is mitigating risks to humans, such as exploring dangerous or hazardous environments. However, traditional robotic systems are bulky, heavy, and too rigid to provide the necessary service. Thus, there is a need for a robotic system that is lightweight, flexible, yet durable enough to meet the needs of the application. This work proposes a robotic platform that utilizes soft actuators but combines them with a traditional rigid body chassis. This platform will use highly flexible and lightweight liquid-based Electroactive Polymer (EAPs) actuators as the actuation mechanism; the unique hydraulic amplification provides large displacements thus acting as artificial muscles allowing the surrounding chassis to mimic the locomotion of a snake. This study will discuss the development and on-going progress of the Hybrid Snake Robotic Platform of our interest. This paper outlines the design for the platform chassis, the locomotion design, the modeling of the dynamic motion, and the fabrication and testing of the platform components. All in all, the development of this platform proposes a robust yet lightweight and compliant robotic platform utilizing a unique form of soft actuation.

Keywords: Soft Robotics, Modular Robots, Grasping
Abstract: Compared with the low adjustment ability of conventional rigid robots in unstructured environments and their poor human-machine interaction, soft robots have been increasingly applied in various fields due to their good adaptability and safe human-machine interaction. Modular soft robots can better adapt to changing production needs and have strong adaptability and versatility. This actuator uses the temporary-bonding-fastener to achieve the assembly of identical soft modules, making the single actuator modules can be quickly and reversibly constructed into soft grippers with multiple configurations. This modular actuator has broad application prospects and provides new strategies to build modular robots.

Keywords: Soft Robotics, Underwater Robotics, Actuation and Actuators
Abstract: Development of underwater robots for unknown underwater exploration has continuously been an attractive research field. However, previous approaches required complicated rigid mechanism to operate the underwater robots. This study introduces the robotic piston combined with inflatable soft pouch actuators based on liquid-vapor phase transition materials, making it applicable to sensitive underwater conditions. A soft pouch actuator made entirely of soft materials was first developed. This soft pouch actuator filled with low boiling point solution operated by electrically Joule-heating, enabling it to be operated silently. The soft robotic piston comprising multiple stacked soft pouches was constructed, which results in reciprocating back and forth linear motion of the piston rod through reversible expansion and contraction motions of the pouch actuator. The dynamic performances of the robotic piston including piston stroke and blocking forces were experimentally investigated. In addition, the soft robotic piston was integrated with the other mechanism as a power source. As a verification of this, the soft robotic piston was utilized for the soft/rigid robotic gripper systems and a walking robot. The results show that the proposed soft robotic piston is well compatible with different mechanism and can adapt to sensitive underwater environment.

Keywords: Telerobotics
Abstract: This paper presents an adjustable compliant control for stable and intuitive teleoperation during various contact tasks. Humans can easily interact with unstructured environment by adjusting force of their arm. However, it is unstable to control a slave robot to interact with unstructured environment only using position and velocity data. Therefore, adjustable compliant control system is necessary to execute various contact task for teleoperation system. In this study, we present the simple adjustable compliant control system using grip force of operator. The relationship between the grip force and compliance of slave robot is based on human’s instinctive skill. Simple experiments were conducted to control slave robot using the proposed system. Results show that operator can modulate stiffness scale of slave robot to control in stiffness motion and compliance motion based on the proposed system.

Keywords: Telerobotics, Medical Robotics and Computer-Integrated Surgery
Abstract: Since the outbreak of COVID-19, healthcare workers are exposed to cross-infection during therapeutic procedures. Non-face-to-face therapeutic process by robotic tele-operation system can reduce the risk of cross-infection and solve the shortage of medical resources such as personal protective equipments. To manipulate therapeutic devices utilized in the intensive care unit (ICU), a compact and multi-DOF end-effector compatible with a small robotic positioning platform is required. In this study, we devise a 3-DOF end-effector that can be applied to a small XY positioner and manipulate a mechanical ventilator. To manipulate a variety of user interfaces of target ventilator (e.g., buttons, knobs, and touchscreen), the device with 3 degrees of freedom is designed for pressing touchscreen and buttons, gripping knob, and rotating it. We apply a soft band-type gripper that enables the end-effector to adaptively grip the knobs different in size and shape. As one of the main working performances influencing response time to medical situation, we evaluate the operational velocity experimentally. As a result, the linear velocity of pressing motion and angular velocity of rotating motion are measured as 61.4 mm/s and 200 deg/s, respectively.

Keywords: Underwater Robotics, Manipulation Planning and Control, Object Recognition
Abstract: Biofouling is a problem of considerable Eco economic significance. Because biofouling brings about skin friction drag of ship and caused disturb ecosystem, the regulation of ship hull cleaning is being strengthened. However, niche area of ship is difficult to cleaning using vehicle-based robot due to the complex cleaning area and the biofouling's fast rate of growth. In this paper, we suggested an autonomous cleaning system using multi degree-of-freedom hydraulic manipulator for niche area of ship hull. Our system used a manipulator to reach a complex niche area and recognized the cleaning target surface using a laser-camera system. Based on the recognized cleaning area, cleaning robot planned an optimal path considering with full coverage and normal vectors of target area. This robot system was verified in ground and underwater, and showed the reasonable cleaning performance.

Keywords: Wheeled Mobile Robots, Object Recognition, Industrial Robots
Abstract: Pipes are used worldwide as a major means of transporting resources. However, defects may occur over time, so periodic inspection is required. Technology for in-pipe robots has been studied to detect defects inside a pipe. The inside of the pipe has irregular surfaces due to rust, corrosion, water, and so on. These factors make the robots out of alignment with pipes. In an unaligned state, control or inspection performance may be negatively affected. This paper proposes a method of estimating the slope of an in-pipe robot inside the pipe using contact sensor modules. The posture of the robot could be estimated using the data of four contact sensor modules in front of the robot and geometric characteristics of the pipe.

Keywords: Wheeled Mobile Robots, Robot Surveillance and Security, Simultaneous Localization and Mapping (SLAM)
Abstract: 일련의 실험을 통해 기존 레이저 기반 위치 인식 알고리즘의 약점을 발견했습니다. 이러한 약점을 보완하기 위해 카메라 이미지의 특징점을 추출하고 단순화하여 일종의 마커로 사용하기로 결정했습니다. 추출된 특징점을 통해 예측된 위치는 레이저 기반 위치 인식 알고리즘과 결합되어 상보적이고 정확한 위치 인식 결과를 도출합니다. 우리는 실제 환경에서의 실험을 통해 보충 된 알고리즘을 확인했습니다.

Keywords: Actuation and Actuators, Biomimetic and Bioinspired Robots, Legged Robots
Abstract: It has been a grand challenge to design the permanent magnet (PM) machine with high torque, high power density. Halbach array, axial flux PM machines are the representative design for reaching the goal. With aforementioned methodologies, however, the cogging torque which proportionally increased with increasing torque density was a major drawback to the performance of the PM machines. Many previous studies have been investigated to reduce the cogging torque, but were limited on the radial flux PM machines and rotors designed with alternative polarity arrangement which has an identical topology as commercial motors. Due to the different topology and the backEMF shape, however, the conventional methodologies for cogging torque reduction are ineffectual to be applied on Halbach array axial flux PM machines, which has significantly high torque density. To address this drawback, this paper suggests an optimal design methodology to reduce cogging torque in Halbach array axial flux PM machine. The methodology was verified through a finite element analysis, and confirmed that the cogging torque, which accounted for 0.2927 Nm of the non-skewed PM machine, decreased to 0.0347 Nm, which corresponds to 88.15%.

Keywords: Actuation and Actuators, Human-Robot Augmentation, Mechanism and Design
Abstract: WalkON Suit IV, a powered exoskeleton for complete paraplegics, has shown its walking speed and ability to overcome obstacles in CYBATHLON 2020. However, its ankle actuators could not generate enough torque to implement the “push-off” motion which is the function of the human ankle during walking to enhance speed and efficiency. A new linear ankle actuator suggested in this paper realized the push-off motion in WalkON Suit IV and achieved a walking speed of 1.05 m/s, which is close to the average of human walking.

Keywords: AI Reasoning Methods for Robotics, Computer Vision and Visual Servoing, Object Recognition
Abstract: This paper proposes a fast and exact k-nearest neighbor search algorithm on hierarchical Nearest Neighbor Graphs (hNNGs). hNNGs combine most of the interesting properties of both Nearest Neighbor Graphs (NNGs) and self- balancing trees. To find the exact nearest neighbor in a hierarchical structure, the concept of bounding spheres of corresponding sub-hierarchies has been applied.

Keywords: Computer Vision and Visual Servoing, Telerobotics, Robotics in Hazardous Applications
Abstract: In order to conduct remote teleoperation tasks safely, it is required to have good visualisation of the remote environment. In practice, 2D cameras are the first option due to their reliability, well-developed compression codecs and transmission protocols. However, 2D imagery fails to provide sense of depth and orientation to operators particularly in highly dexterous configurations, resulting in long training sessions with slow learning curves. A potential solution consists in enhancing visuals with 3D information using 3D cameras, however, real-time transmission of large, high-resolution point cloud data at high frame rates remains a challenge and a burden to the network. This manuscript presents ongoing work applying two different approaches to prioritise transmission of certain parts and components of the scene in combination with tree-based compression structures to lower bandwidth needs aiming at real-time 3D data remote visualisation.

Keywords: Contact: Modeling, Sensing and Control, Force and Tactile Sensing
Abstract: Probe-based confocal laser endomicroscopy allows optical biopsy with small and flexible imaging probes. To obtain microscopic images in focus using the imaging probe, it is necessary to maintain the imaging probe within depth of focus between the probe and tissue. In the paper, we propose a three-dimensional force sensor based on fiber Bragg gratings for probe-based confocal laser endomicroscopy. The dual structure of the proposed sensor improves stiffness balance, which brings comparable axial and transverse resolutions. From finite element analysis (FEA), the dual structure increases the resolution of transverse force 4.6 times compared to a single structure. The dual structure design enhances the ratio of transverse force to axial force—1.2, whereas it is 7.5 for the single structure. For the transverse and axial forces, FEA results show resolutions of 2.949 mN and 2.398 mN and sensing ranges of 7.4 N and 1.8 N, respectively.

Keywords: Dynamics and Control, Manipulation Planning and Control
Abstract: This paper proposes a practical auto-tuning Proportional-Integral-Derivative (PID) control robust against significant internal and external dynamics changes for robot manipulators. The proposed PID control achieves robustness against significant internal and external dynamics with a modified Nussbaum class auto-tuning control method. The proposed PID control is practical as 1) there are only two tuning parameters, 2) each parameter has a clear meaning, and 3) the proposed PID control does not require a robot dynamics model. With the adoption of the forgetting factor, the proposed PID control prevents drift in the adaptive gain caused by the numerical integration of an update parameter accompanying sensor quantization error and noise. The experiment with a two degrees-of-freedom direct-drive robot verified the robustness of the proposed control against significant internal dynamics changes (4kg payload, 110% of robot arm mass) and external dynamics change due to stiff spring (180N/m spring). The constant PID tuned with the payload was unstable when there was no payload. These indicate that, under the significant robot and external dynamics changes, the proposed auto-tuning PID control is robust and can guarantee stability.

Keywords: Force and Tactile Sensing, Medical Robotics and Computer-Integrated Surgery
Abstract: Supermicrosurgery, which anastomize blood or lymphatic vessels and nerves with very small diameter, requires high precise motion. With the advantages of surgical robot, robot-assisted supermicrosurgery showed its feasibilities through clinical trials. However, force sensors, which are significant consideration in surgical robot, are still absent. Placing force sensors directly at forceps jaw provides most precise measurement, however there are more stringent design requirements to satisfy. FBG sensors are great solution for above problems, however measuring 3-DOF forces with high axial sensitivity and balanced sensitivity is difficult. In this paper, we present a 3-DOF force-sensing forceps with double compound parallelogram flexure and four FBG sensors to accomplish the high axial resolution with great balance along all axis. Sensing principel has been derived, and performance of designed sensor is validated through FEA. Simulation results shows great decoupled and balanced high sensitivity in all axis.

Keywords: Grasping, Manipulation Planning and Control, Dynamics and Control
Abstract: Reinforcement Learning(RL) is a method capable of solving tasks in complicated environments. Due to this capability, RL have been considered as a new method to control robots. However, in order to RL to learn to solve complex task, numerous trajectory data are needed. Therefore, improving the data efficiency is one of the main issues of RL. To solve this issue, our model uses a novel method called ‘Backward Dynamics’ to enhance training. Unlike typical model-based RL, which focuses on predicting future states, our approach infers possible past states and uses it to generate virtual trajectory. These trajectories are used during training procedure, improving the learning speed.

Keywords: Industrial Robots, Dynamics and Control, Wheeled Mobile Robots
Abstract: In this paper, a self-driving vehicle platform was developed by modifying an electric vehicle for a golf cart. The self-driving electric vehicle is equipped with LiDAR sensor, depth camera, and ultrasonic sensors. An electronic power steering system was applied to steer the autonomous platform. Building a map in a warehouse and autonomous driving performance experiments were conducted through the developed autonomous driving platform vehicle.

Keywords: Manipulation Planning and Control, Industrial Robots, Dynamics and Control
Abstract: In order to implement acceleration trajectory planning in the task space, it’s typically required the computation of the time derivative of the Jacobian matrix. This can be expressed as the product of the Jacobian derivative for the joint variable and the joint acceleration vector using the chain rule of differentiation. The numerical method is not suitable for real-time control since it accompanies high computation costs and does not guarantee reliable calculated values at all times. In this paper, we propose a geometrical representation of the time derivative Jacobian for fast and reliable computation. The geometric analysis is presented according to the Jacobian reference frame for an n-DOF serial manipulator.

Keywords: Manipulation Planning and Control, Object Recognition, AI Reasoning Methods for Robotics
Abstract: We propose a Superquadric-based Interaction Network (SI-Net) for learning to predict the motions of the objects after a pushing action is applied to the environment, where the objects' information and physical quantities such as friction coefficient are unknown. SI-Net first recognizes the unknown objects' shapes from raw vision data using superquadrics. Then it predicts the motions of the recognized superquadric objects after the pushing action is applied to the environment. Our newly proposed method predicts motions on a lower-dimensional object representation, the neural network can be trained data-efficiently and easily applied to real-world pushing manipulation tasks such as moving and singulating the target object. SI-Net shows remarkable performance on the motion prediction tasks and real-world manipulation tasks.

Keywords: Mechanism and Design
Abstract: This proposes a self-alignment joint mechanism using the tensegrity structure. The center of rotation of the human joint varies every moment during motion, which makes users of the robotic exoskeleton uncomfortable. To alleviate this issue, a self-alignment joint is presented using the tensegrity structure. The proposed joint mechanism does not provide a constant rotation, but rather makes an angular difference between the thigh and calf to assist the user.

Keywords: Mechanism and Design, Grasping, Robotic Hands
Abstract: The precision grasping ability for various shape objects depending on the direction of approach have to adapt to the provided candidate grasp set, i.e., triplet. Therefore, the robotic gripper should adapt to the target object’s pose and shape depending on the approaching direction. In this paper, a three-fingered robotic gripper with various sizes and shapes of triplet adaptability was proposed. In prior research, the kinematic topology is analyzed by simulating synthesized kinematic topology composed of revolute and prismatic joints under the force closure condition. Based on this, We propose the radial slider-crank mechanism, which can change palm-size and lateral arrangement among fingers.

Keywords: Mechanism and Design, Industrial Robots
Abstract: Harmonic drive is used for industrial robots and precision mechanical equipment. It is characterized by its light weight. Since flexspline which is one of the main parts of harmonic drive is constantly deformed, stiffness in the radial direction and circumferential direction is required. In this paper, the torsional stiffness of flexspline which is necessary for accurate rotational motion of harmonic drive was analyzed. A finite element model was constructed based on the torsional stiffness experiment conditions. Torsional stiffness was confirmed by applying three rated torques, and the finite element model was verified by comparison with the experiment.

Keywords: Mechanism and Design, Physical and Cognitive Human-Robot Interaction, Rehabilitation and Healthcare Robotics
Abstract: Nowadays, the number of people with gait disorder as one of post-stroke symptom is gradually, and continuously increasing. For these people to have similar Activities of Daily Living (ADL) as people without gait disorder, active knee assistance is essential. This requirement is coming to the surface. The philosophy of assisting robots this paper presents is that assistant robots should only help people, while not disturbing the wearers’ natural movement. For this philosophy, robot hardware has to be well-constructed before application of any kind of control algorithm. Therefore, for flexion angle from 0 to 90 degrees, knee assistant robot hardware is proposed to secure natural movement of patellar and while minimizing force between tibia and femur in this paper.

Keywords: Mechanism and Design, Soft Robotics
Abstract: In this paper, we present a novel Multi Pulley Module(MPM) for soft wearable robots that assist the standing and walking of patients. Compared to previous approach of soft wearable robots, MPM is adaptable to various gait patterns. Robots equipped with MPM provide a more versatile assistance in that their dynamic models can easily be altered. Users can get customized assistance according to their gait pattern by simply attaching and detaching pulleys. For the verification of the performance, a pilot study was conducted which investigated the effect of the hip flexion/abduction assistance for a static pose of the subject. By comparing the electromyogram with surface elctromyography(sEMG), it was shown that MPM effectively reduces the activation of target muscles by 35%, 14.2%, respectively. This preliminary study demonstrates that MPM can help multiple patients while greatly reducing the effort of part replacement.

Keywords: Multi-Robot Systems, Motion Planning and Obstacle Avoidance, Industrial Robots
Abstract: In this work, we propose a modification of a multi-agent pathfinding (MAPF) algorithm based on conflict-based search (CBS). CBS and its variants perform searches to generate collision-free paths of multiple agents by defining conflict situations (i.e., collisions) to be avoided. Often, existing methods do not concern about the physical volume of the agents whereas autonomous mobile robots (AMRs) operating in warehouses or smart factories often have bulky volumes. The difference in the assumption about the bodies of agents is one of the hurdles to use the MAPF methods for real-world multi-robot applications.

Our approach considers those realistic robot volumes and shapes in the CBS method by defining additional conflict situations that robots collide owing to the enlarged body volumes. Our experiments validate that the newly defined conflicts enable the robots to avoid collisions even in a packed environment where robots move very closely. We show extensive tests and comparisons to show the improvement in the number of successful navigation without collisions. We achieve 100% of the success rate in finding collision-free paths for all robots while maintaining the path costs almost the same. In return, we sacrifice the computation time to find paths.

Keywords: Object Recognition, Computer Vision and Visual Servoing
Abstract: Robots make use of RGB-D cameras which provide RGB images as well as depth information as data sources for various visual processing algorithms. In particular, RGB and depth images can be combined into point cloud data for advanced applications. Due to the interference of ambient light and the limitation of the transmission medium, however, the size and clarity of the data are insufficient. On the other hand, GAN-based super-resolution methods have made great progress in the field of image restoration. In this submission, an image resolution enhancement approach is proposed to increase both the size and the quality of RGB and depth images, while reducing the influence of illumination. For this purpose, we use the Restormer to denoise the RGB/depth images and then use Real-ESRGAN to increase the pixels of RGB/depth images. Through this work, the quality of the RGB/depth images is significantly improved.

Keywords: Rehabilitation and Healthcare Robotics
Abstract: Gait analysis in a medical environment can provide important information to check the recovery and rehabilitation status of patients in healthcare. In this paper, we propose a method to detect the user's gait and measure the number of steps and distance by using a 2D laser sensor mounted on an interactive robot. Specifically, a leg detector trained with a random forest was applied to the user's leg detection, and the error for a given gait experiment was verified and the utility in various environments was analyzed.

Keywords: Rehabilitation and Healthcare Robotics, Mechanism and Design
Abstract: Treatments for ADHD symptoms lack the resources needed to continue to be successful and are outdated in many ways. Several studies employing Virtual Reality to treat ADHD symptoms have been undertaken, however they have limitations because they primarily focus on improving a patient's ability to focus while failing to address other deficiencies, such as social competence. Our study will develop and apply a 3D avatar trainer in a virtual 3D gym to reinforce attributes including concentration, restraint, physical perception, auditory skills, sequencing skills, and judgment abilities through a series of exercises with the avatar. We conducted a pilot experiment and intend to prove the effectiveness of treatment for ADHD patients by using virtual reality in comparison to canonical methods.

Keywords: Rehabilitation and Healthcare Robotics, Modeling, Identification, Calibration, Dynamics and Control
Abstract: It is crucial to quantify upper limb mechanical impedance because it provides a wealth of information ranging from basic muscle properties to assessing the stages of neurological impairment like stroke. The recent stochastic estimation method of the mechanical impedance has advantages over other methods and needs one assumption to be met: the upper limb behaves linearly. However, a previous study using a low friction robot (i.e., MIT-MANUS) showed low partial coherence (<0.8), indicating that the upper limb may behave nonlinearly. A recent study elucidated that the low coherence could be due to nonlinear joint friction in the robot rather than the upper limb using a robot with large friction due to the use of harmonic drive with a large velocity reduction ratio. Thus, there is a strong need to investigate further if the low coherence is due to the nonlinearity in the upper limb or the robot. To clarify the effect of uncompensated robot friction, the experiment was carried out using a direct-drive robot almost identical to the MIT-MANUS, and the spring arrays, the second-order linear system. The finding revealed that if the friction is not compensated, it lowers the coherence functions. This is indirect but significant proof that the cause of low reliability stemmed from the friction of the robot, not the human arm.

Keywords: Robotics in Hazardous Applications, Motion Planning and Obstacle Avoidance, Computer Vision and Visual Servoing
Abstract: Autonomous mobile robots are expected to be applied in challenging environments such as snow-covered environments. In a snow-covered environment, it is difficult to detect pavements on snow-covered road surfaces by semantic segmentation. To overcome this problem, we present a novel method to improve the pavement detection accuracy of semantic segmentation by using a generative adversarial network (GAN). Our results show improved pavement detection accuracy. Therefore, we can control the motion of the mobile robot along the detected pavement.

Keywords: Simultaneous Localization and Mapping (SLAM), Intelligent Robotic Vehicles, Motion Planning and Obstacle Avoidance
Abstract: In this paper, the concept of a mobility scooter capable of autonomous driving is introduced and the functions to be developed are explained. The developed mobility scooter is equipped with an autonomous driving function to prevent accidents by recognizing indoor and outdoor environments by applying various sensor systems. In this paper, elemental technologies developed for indoor and outdoor autonomous driving of mobility scooters are explained, and their functions are tested indoors to show their effectiveness. In the early stages of development, we plan to develop various technologies for user safety as well as indoor and outdoor autonomous driving technologies and apply them to the field quickly.

Keywords: Social and Socially Assistive Robotics
Abstract: This paper introduces an integrated development environment (IDE) that allows end-users, who aren't usually programming professionals, to set up their own interaction scenarios with a social robot. This IDE allows users to edit, save, and share generated scenarios with other users, alongside providing a virtual environment to execute scenarios without requiring a physical robot. This IDE's editing software provides a user interface centered around connecting function blocks, much like beginner-friendly visual programming tools. Given that these building blocks enable high-level AI functions, subsequent interactions within the programmed scenario can accommodate flexible yet sophisticated AI perception and expression. This includes functions that classify natural language expressions as positive or negative responses to binary questions, or specific answers to multiple-choice questions; recognize gender, emotion, and pose; automatically generate facial expressions and gestures to match speech output according to sentiment analysis; etc.. In the future, we aim to enhance interfaces within this software with 3rd party intelligent functions and share findings with the broader robotics community.

Keywords: Social and Socially Assistive Robotics, Physical and Cognitive Human-Robot Interaction, Object Recognition
Abstract: This paper proposes a methodology for a mobile robot to automatically set an actual movable destination from the user's abstract command for object delivery in an indoor environment. To this end, the robot constructs a topological map for each node through GVG (Generalized Voronoi Graph) fine-tuning and geometric operation from the map information of the space, and maps abstract space information from object recognition for the space while searching for nodes for each divided area. This mapping information allows the robot to search for a node to the destination of an optimal location for final product delivery from the geometrical correlation with the grid map.

Keywords: Soft Robotics, Grasping
Abstract: Soft wearable hand robots with tendon-sheath mechanisms are developed for aiding people who have lost their hand function. For these robots, knowing the fingertip force and grasping objects’ sizes can successfully grasp objects without wasting force or inordinate deformation of the objects. As an approach, attaching the additional sensors to the wearing part restrict the behaviors of the user and the sensor can cause electrical risks such as sensors failure. In this study, we developed a prototype of a robot system to develop a learning algorithm which can estimate grasping object sizes without any sensors on the glove. It is important for the system to acquire precise data considering nonlinear characteristics of the tendon-sheath mechanism, finger, and objects. In the prototype, a learning-based algorithm based on LSTM is included to estimate grasping objects’ size. The overall system architecture is also presented.