The problem of the manipulator avoiding interference with obstacles in the work environment and planning a safe route is one of the big problems in robotics automation. As one of route planning with obstacle avoidance, RRT aims at speeding up the calculation by using random sampling. RRT has the advantage of high obstacle avoidance capability. In recent years, researches on a robot vision system combining robot and vision progressed. In the obstacle avoidance route plan, information on the obstacle position from a camera has advantages such as easy calculation. The purpose of this research is to calculate the position of an obstacle necessary for route planning using a 3D camera and propose a method in which the manipulator moves to the target position while avoiding the obstacle by the improved RRT route planning method.

Robots are expected to be used as service robots or nursing care robots. Collision mitigation technologies are very important for those kinds of robots, because they are afraid to have collisions with humans. To cover robots with buffer materials and to use elastic actuators are effective to keep safe. And collision mitigation control is also important because to keep a position control will not be the best way to keep safe. In this paper, collision mitigation control law to make the impact force smaller is discussed. To give the maximum opposite force may be the best way, but a reaction movement becomes very large. The initial value compensation is applied to suppress the reaction movement and confirmed its effectiveness.

In this paper, we propose a model-based reinforcement learning framework combining Model Predictive Path Integral (MPPI) with a Deep Path-cost Predictor that outputs a state-trajectory cost given an image sequence and a control input sequence as input. We validate the effectiveness of the proposed method by carrying out 2DOF robot arm reaching tasks with multiple targets in simulation.

This paper proposes a novel motion planning method in rough terrain that improves our previous work. The previous method can generate a collision-free path within a short time, however, there is room to improve the quality of its path, especially about the length of its path. In this work, we add a path deformation step after a path generation step. The path deformation is performed by optimization. The objective function of optimization consists of path length, difference between original path and deformed path, and risk of deformed path. Experimental results show that the proposed method can deform a re-planned path appropriately and can reduce the length of the path.

The master-slave system with time delay causes overshoot of the position response of the slave arm, so it is difficult to apply it to applications requiring precise operation such as remote surgery. In this paper, we propose a master trajectory predictor to reduce the influence of time delay and to suppress overshoot of position. We use LSTM model which is one of machine learning algorithms, to predict the position of the master. We report the results of the trajectory prediction using the proposed method for using a degree of freedom master device driven by pneumatic cylinder.

Currently, a driving mechanism with multiple degrees of freedom is required. The purpose of this research is to design and prototype a basic experimental apparatus which can rotate a sphere in 3 degrees of freedom by using active-casters, and to develop a system for controlling the motion of a ball. As a preliminary step of using the actual machine, simulations were conducted using 3D CAD in which three casters were arranged around the ball so that each steering axis in rotatable along the x, y, z axes with the center of the ball as the origin. First, I derive relational expressions between balls and casters. Next, I confirmed that the derived formula can accurately represent the relationship between the ball and the casters. Finally, we confirmed that the ball can be rotated with the target axis of rotation and angular velocity using three casters.

The trajectory planning of assembly tasks, which demand high precision, such as peg-in-hole task, requires consideration on errors driven from robot control and the environmental model. This research proposes a trajectory planning method in peg-in-hole task, taking positively advantage of the trail of force control. We confirmed that the proposed method realized the peg-in-hole task. In addition, admittance control succeeded in the task in smaller contact force than position control.

Collecting the damage situation is required when large-scale disasters occur. Even in places where humans can not enter, information can be collected by using unmanned aerial vehicles(UAV). Because of the past research, it is possible to grasp the disaster situation efficiently by creating a database with multiple resolutions of large-scale terrain data. The database needs updating, but it is not efficient if the whole database is rebuild. In this paper, we focus on the multiresolution database of the large scale terrain data and propose an efficient updating method of it by using UAV and ROS.

This paper describes a real-time 3D measurement technique using multiple global navigation satellite system (GNSS) receivers and antennas on a small unmanned aerial vehicle (UAV). At disaster sites, rapid 3D measurement is highly required. However, in our previous research, it took time to generate a 3D map because the computation process after a flight of the UAV was needed. In addition, this technique needs 30 position calculations to estimate the UAV attitude. Accordingly, it is difficult to estimate the attitude using an onboard computer in real time. Therefore, we propose a method using six GNSS antenna positions to estimate the UAV attitude. Because the proposed method can reduce position calculations, it can be used for real-time UAV attitude estimation. Then in the proposed system, we use robot operating system (ROS) to build distributed system. By using the proposed system, we achieved real-time 3D measurement in 151.7 mm horizontal root mean square (RMS) error and 4.8 mm vertical RMS error.

To realize aerial manipulation using a drone, it is required to control the position of the center of gravity. In order to fly safely, it is necessary to obtain the parameters used for the control before takeoff. However, it is not easy to determine the parameters in advance due to changes in parts or batteries or changes in the objects to be transported. Therefore, in this research, we propose a method to estimate these parameters before flight and show its effectiveness by experiment.

Multicopter-typed unmanned aerial vehicles, called drones, have been attracting wide attentions for hobby, aerial photography and various industrial use. For industrial use, drones usually fly in areas with various aerodynamic disturbances and obstacles, and it is essential to control its attitude and position accurately for save operations. Flying animals, such as insects and birds, fly with flexible wing and achieves highly robust and efficient flight. In this study, inspired by unique morphology and structure of animals, we have developed high-performance flexible rotary wings for drones. Gust response of propellers are measured experimentally using 6-axis loadcell. Passive deformation under the disturbances are filmed by a high-speed camera. Through the experiments, it is found that the insect-inspired flexible propeller can suppress the unstable torque under gust by the passive wing deformation.

The authors proposed a novel hose type robot, which can fly directly into the fire source via a water-jet and demonstrated that a robot with a length of approximately 1.8 m can fly the air by leveraging the water jet.That named Dragon Fire Fiter (DFF). However, DFF had some problems. Most important problem is that The lack of control input of the robot could not suppress several modes of body oscillation. In this study, the authors propose the passive mechanism using the wire to solve that problem．The authors also model the mechanism to evaluate the proposed mechanism.

This paper presents a monaural speech enhancement method for a hose-shaped rescue robot based on a deep speech prior. Speech enhancement is crucial to make a robot operator succeed in detecting human voices because audio signals captured by a microphone on the robot are contaminated by ego-noise. We have been developed three enhancement methods: 1) a blind speech enhancement called robust nonnegative matrix factorization (RNMF), 2) an extension of RNMF with a pre-trained noise model, and 3) another extension of RNMF with a deep speech prior, i.e., a pre-trained speech model based on deep learning. In this paper, we develop a new extension of RNMF by combining the pre-trained noise and speech models as a unified model and evaluated these methods on a hose-shaped rescue robot whose ego-noise consists of vibration-motor and air-jet noise. Experimental results show that the new method outperforms the three RNMF methods when the signal-to-noise ratio is equal to or less than +5 dB.

We propose a planar omnidirectional crawler mobile mechanism for the mobile basis of search and rescue robots. One of the advantages of the planar omnidirectional crawler mechanism is that it allows the robot to move in any direction without causing large slippage on soft ground because it does not require a turning motion when the robot switches its traveling direction. This study realized a configuration with two left and right unit crawlers that enables a turning motion. By using this prototype machine, we investigated how the turning and translational motions of the crawler mechanism exert influence on the imitating soft ground by measuring the flow of sand on the horizontal surface. As a result, the translational motion was quantitatively shown to switch the travel direction with less degree of disturbance to the road surface than the turning motion.

Biological snakes, with an elongated, legless and lithe body, can adapt to complex and varied environments. Snake robots, which are developed by imitating the characteristics of biological snakes, can work in various circumstances, such as a disaster site and large scale plant. In this research, we propose new locomotion that snake robots could move in crowded pipes by thread motion. In this paper, we present a basic idea of the thread motion. We also conduct some experiments to verify the effective of the motion.

A snake robot is expected to be used for searching narrow space in disaster sites and inspecting ducts and pipes on daily use. In the previous study, bending helical rolling motion is proposed as a method for moving a snake robot in a bent pipe. In the actual pipes in the plant, not only straight pipes and bent pipes but also branch pipes exist. A snake robot need to be able to move branch pipes in order to move freely pipes in a plant. In this paper, bending helical rolling motion is used as a method for moving branch pipes.

Snake robot can change its own shape flexibly, so it is expected that a snake robot investigates ducts and pipes more effectively. In previous research, a snake robot can move in pipes using helical rolling motion. However, it is difficult that snake robot checks inside of pipes by using camera image because the position and direction of its head link is not stabilized. Therefore, we adopt Virtual Chassis and inverse kinematics to stabilize the head link of snake robot when the snake robot is moving in helical rolling motion. Virtual Chassis is a method of defining center of gravity as an origin of coordinate system, obtaining the spatial orientation of a snake robot by the principal component analysis, and pasting a provisional coordinate system to a snake robot. We compared the case with the head link control to the case without the head link control, and verified how much the camera image has changed. From the experimental results, the center of the camera image approaches to the center of the pipe by the head link control.

In this research, we installed a sensor capable of measuring the pressure around the full circumference on a snake robot and developed behavior of the snake robot whose joint movement changes according to the measured value of the pressure sensor. We propose an algorithm that a snake robot push environment in the case where propulsive force can be obtained against pressure received from the environment during movement, and detach from the environment when propulsive force can not be obtained. Experiments were conducted to evaluate the proposed method. In the experiment, the snake robot moves in a test course. Time and energy consumed by moving in the test course were compared between when using the proposed method and when not using it.

The paper presents an interface system which locally changes a posture of an articulated mobile robot. In the system, we assign two points in the robot’s body, one is the base point, and the other is the controlled point. In addition, we set the position of the target point in the system. The robot can locally change its form by controlling the joints as a redundant manipulator. The effectiveness of the proposed system is demonstrated by using a physical simulator.

This paper describes visualization of a snake-like robot and its surrounding environment for teleoperation support. For visualization of the robot moving in the pipe, we propose the display of a self-position of the snake-like robot and piping shape. The self-position of the robot is estimated by introducing the notion of odometry to the snake-like robot. The piping shape is shown by drawing contact points obtained by the contact sensor mounted on the robot. For visualization of a snake-like robot moving in a large space, we propose the display of the surrounding environment with a laser sensor. The snake-like robot scans the environment three-dimensionally by swinging a head unit which has the laser sensor. As a result of visualizations, The robot moving in piping correctly shows the self-position of the robot and the state of the surrounding piping of the robot, therefore, it is effective for teleoperation support. In the case of the robot moving in a large space can visualize some objects in the environment, however, the obtained point clouds by the sensor were inclined because of the effect of the localization error of the robot.

We aim to develop a method to measure the height of sedimentary sand from dog's moving trajectory and ground shape data. In sediment-related disasters, it is important to grasp the height of sedimentary sand for the search of victims. If we can visualize the search trajectory of rescue dog three-dimensionally, we can measure the sedimentary sand accumulated place and the depth of it comparing with the topographic map before the disaster. Therefore, we estimate the search trajectory of SAR dogs in three dimensions using the sensors, such as IMU and GNSS, which is mounted on dog. From the result of experiment, we can estimate the height of sedimentary sand from dog's running 3D trajectory.

Understanding the topography of a disaster site will greatly improve the efficiency and safety of a search and rescue mission, in which, the best representative of these information is a 3D map. The purpose of this research is to develop a canine suit with 3D mapping and localization capability to collect topographic information along with search and rescue canine’s operation. The suit is suitable to operate in both indoor and outdoor environment and is equipped with 3D LiDAR sensor, an IMU sensor and a GNSS receiver. It has evaluated registration error of about ±3 cm. Its weight is 2.9 kg and can be put on dog over 30kg. It is also implemented with LOAM SLAM module to create robust 3D map as well as to handle point cloud registration error due to the rapid movement of canine.

This paper describes the cyber-enhanced canine suit No. 7 revision 2 (called No. 7_r2) that improves overheating issue and suit inclination issue of the canine suit No. 7. Overheating issue was handled by applying heat countermeasures on WiFi dongle and camera. Suit inclination problem on dog was solved with the adjustment of chest belt length. Also, the suit was made lighter and smaller to reduce the burden for dogs. The weight of the suit was revised from 1.35 kg to 1.08 kg, and the size was reduced by 2-3 cm in length and breadth. With these revisions of No.7 suit, we started lending the suits to 2 rescue canine associations to be ready for next disaster.

In this paper, we developed a guidance suit for a dog with light sources. The dog is guided by spot light sources which can illuminate in three directions including frontal, left and right according to remote controller. First, we allowed the operator to easily adjust the irradiation angle of the light sources. Then, the mounting position of light sources are examined. Two radiators were mounted on the shoulder of the dog to irradiate in frontal direction as to prevent being blocked by dog’s head, while left and right radiators were mounted on the side of the suit to easily guide the dog left and right. A board PC was also equipped with the suit so that radiators can be controlled by remote controller. Additionally, sensors including camera, GNSS and IMU were installed to monitor dog’s action from a remote location. In verification of guidance, we succeeded in navigating the dog.

ImPACT Tough Robotics Challenge is a national project of Japan Cabinet Office (2014-18) that focused on tough technologies of robotics to give solutions to disaster response, recovery and preparedness. It consists of subprojects of six types of robot platforms and several component technologies integrated with the robots. Cyber rescue canine suits for monitoring dog’s behavior and commanding her to move has shown effectiveness at regular exercise with Japan Rescue Dog Association. A new serpentine robot, Active Scope Camera, that can crawl and levitate in gaps of a few cm to search in rubble piles. Structural assessment and radiation measurement were performed by Active Scope Camera in Fukushima-Daiichi in December 2016 – February 2017. High mobility of serpentine robots in ducts, in/out of pipes, on uneven terrain, and on vertical ladder was shown as well as climbing a step 1 m high by a robot body 1.7 m long. Omni Gripper can grasp wide variety of targets without positioning by jamming phenomenon. Robust flight of Mini-Surveyor under difficult condition contributed response to Northern Kyushu Heavy Rain Disaster by gathering high-resolution image in inaccessible area in July 2017. WAREC-1 can move by 4 legs, 2 legs, and crawling and can climb vertical ladders. A double-swing dual-arm mechanism, operator support by bird’s-eye view images, and that by sense of force and touch enabled a construction robot to perform both high-power and precise remote tasks. Field Evaluation Forum displayed these achievements in disaster scenarios.

In this paper, I propose a passive gravity compensation mechanism using double-angle formula. Several type of gravity compensation mechanism using trigonometric function have been studied. Without using approximation by cam or friction, these gravity compensation are independent of the link angle. In this research, I try to improve the degree of freedom of arrangement of springs used for compensation, by using double angle formula. The theoretical design of mechanism and examination by trial production were conducted.

The off-road vehicle with a crawler and a movable leg mechanism can travel on uneven ground. To carry heavy object stably, large weight and size bodies are required for theses robots. However, that reduces the scope and time of activity. This paper proposes the use of the center of gravity movement mechanism. In previous studies, a serial linkage mechanism moves the center of gravity. However, the movable direction is limited to the anteroposterior direction. In this study, to travel rough terrain, the center of gravity moves not only in the front-rear direction but also in the left-right direction. To reduce the load on each actuator compared with the serial linkage mechanism, the proposed robot employs the coupled drive system. We performed experiment, and confirmed the movable range of the proposed mechanisms.

We have been studying noncontact object manipulation technology in which a single ball-shaped object is floated and controlled for its 3D position with multiple air jets driven by a pan-tilt actuator. In this paper, we try to control position and orientation of an arbitrary shaped object. Here an arbitrary object is connected with a triangle platform which is composed of three spheres linked with thin wires. Each sphere is spatially controlled by an air jet unit which consists of an air jet on a pan-tilt actuator. Kinematics of the air jet platform as a parallel link mechanism is calculated and a control method for the air jet platform is proposed.

Various kinds of load test are conducted in the development of mechatronics devices. Mechanical single axis spring and hydraulic devices are often used as a load force generator. However, it is difficult to apply this test method to complex mechanical system such as steering mechanism of vehicles. Against this background, authors have proposed the method combining Stewart platform and impedance control to deal with this problem. Combination of Stewart platform and impedance control would realize multi-axis spring. It is expected that this multi-axis spring is useful to solve above mentioned problem. This paper describes the development of the mechanical impedance model, implementation of the impedance control system. After that evaluation of the control system by experiments is shown. From these experiments, it was confirmed measured spring constants agree with spring constant specified by the impedance model.

A novel KARAKURI moving core for injection molding which enables the core to set to and release from complicated moldings is proposed. The KARACURI moving core is comprised of cam mechanism and parallel mechanism. The cam is driven by the injection molding machine so as to synchronize with opening and closing the mold. Followers of the cam are connected to the moving parts of the parallel mechanism. A core attached at the end-effector of the parallel mechanism is driven by the parallel mechanism synchronized with opening and closing the mold. In this paper, conceptual design and design procedures of the KARAKURI moving core is introduced. Two-dof KARAKURI moving core with translation cam was designed as a simple example.

Forward/inverse kinematics solvers for a 3-UU parallel mechanism are provided in this paper. The mechanism can tilt largely in two directions and support heavy load, so that it is available for an inclining platform. In spite of that, its kinematics cannot be solved in straightforward ways due to an interesting feature that its degree-of-freedom in theory is zero. Although an iterative method was proposed in the previous report, its slow convergence had been a problem. An artificial neural network is used for the solver, where the training data is prepared by the above method. The network shows a fast reasoning ability that is available for real-time control.

In order to achieve tasks in the real world environment, humanoid robots have motors and reduction drives optimized in relation to weight and size for providing the necessary torque and angle speed. Therefore, having torque or angle speed outside of the predicted range will usually cause the task to fail. In this research, we propose a joint module with a two-stage transmission mechanism during driving and a joint locking mechanism during non-driving so that the appropriate torque and joint speed can be attained during the task. By applying the joint module to a tricycle type robot and switching the driving state during the task execution, we were able to both reduce the motor load when lifting heavy objects at driving time and keep high rigidity of the joint at non-driving time.

Posture optimization and compliance control for the kinematic redundant three-dof planar parallel robots with rotational mechanisms embedded in the moving part is proposed. The four-limbs controls the positions and angle of the moving part frame, as well as the angle of the rotational mechanism inside the moving part. There are several mechanisms for the embedded rotational mechanisms such as pulley, gear train, rack-and-pinion and crank. Integrated modeling that convers the combinations of the rotational mechanisms and type of the limbs was proposed in the previous paper. This paper discusses the cost function for the posture optimization of the robot in detail, then relates the result to newly derived compliance control of the parallel robot.

This paper described that the ankle joint measurement and control device which has six degrees of freedom by a parallel link mechanism with six actuators. The composite motion of the ankle joint was reproduced in the developped device, and the effect of the combined motion on the lower leg muscle was measured by moving the ankle joint along the reproduced trajectory. The myoelectricity of Tibialis Anterior (TA), Soleus (SOL), Medial Gastrocnemius (MG) and Lateral Gastrocnemius (LG) of the motion combined inversion and plantar flexion was close to that of plantar flexion. The myoelectricity of Peroneus Longus (PL) was close to that of inversion. The results suggested the possibility of local rehabilitation of the lower leg muscle by the device.

In order to grasp and hold uneven objects, a robust adhesion mechanism is required. This paper propose a new universal hand which has controllable adhesion using Magneto-Rheological Elastomer (MRE) as a technique to grasp to non-magnetic materials and uneven surface. This material is similar to Magneto-Rheological Fluid, however it is both simple to use for rough surface and clean to sticking surface. In this paper, we propose a composition of MRE for robot hand and experimentally evaluate yield stresses in a normal direction. Based on the MRE, controllable adhesion using MRE is applied to various type of material and shows effectiveness for uneven surface.

This paper shows manipulation control by a two-fingered robotic hand including 100 ms delay within the controller. First, we show the mechanism of the robotic hand having twist-drive actuator for joint actuation. This robot is able to control extremely small fingertip force according to 3 Hz sinusoidal reference trajectory. Finally, we show successful grasping and manipulating motions in the basis of large and small grasping forces during maintaining object orientation.

Able to grasp objects of any shape and size, universal jamming grippers using magneto-rheological fluid have been developed. However, as they require magnetic flux to change the stiffness, conventional MR grippers consume electricity to attach and detach a permanent magnet, to drive a solenoid of an electromagnet, and to cancel the flux of an electro-permanent magnet. In this research, the authors proposed a new method of supplying the flux using the Internally-Balanced Magnetic Unit, a mechanism that enables a permanent magnet to be controlled by a much smaller force than its attraction force. They realized its prototyping model and conducted experiments to show its effectiveness.

In recent years, industrial robots have been applied to food production because of its huge potential market. Although cooking large amount of foods are done by specially designed cooking machines, dishing up the cooked foods remains as a labor-intensive task. To solve this problem, we propose a new food handling robot hand composed by soft actuators which can handle easily damaged natural food and convey it at high speed. In this paper, characteristic of the soft actuator was shown and bending mechanism was analyzed by FEM. In addition, the proposed gripper was evaluated by conveying experiments with some food samples.

Soft robot hands can stably grip many kinds of objects. However, it is difficult to manipulation requiring slip. Therefore, in a previous study, we developed friction change mechanism utilizing the transformation of contact area by structural deformation. We call this mechanism contact area variable surface (CAVS). In this paper, we analyze the factors that CAVS functions by modeling, and consider how to use it practically. In order to use CAVS in manipulation, we observe the deformation inside the CAVS with a camera to judge whether it is line contact or surface contact. Through the operation of grasping and placing an object in an accurate position, we verify the validity of CAVS.

We are developing robot hands with proximity sensors capable of acquiring information on the vicinity of an object. Up to now, a grasp posture at which the fingertip is facing directly to the object is considered and network structure proximity sensors have been designed for the grasping. By using the proximity sensors, it is difficult to realize a grasping at an arbitrary position of the fingertip, which is necessary for complicated grasping tasks. In this paper, a network structure proximity sensor is designed to meet the need and basic characteristics are investigated by experiments using a prototype of the sensor.

Caging is a method of object constraining based on geometric information. Compared with conventional grasping, caging requires less information and simpler algorithm, but it cannot determine the position and posture of the object in the hands of robots. Our group proposed a new grasping method that takes advantage of caging called “caging-based grasping.” Until now, rigid objects were mainly considered as the target of caging-based grasping. There is, however, much demand for grasping various deformable objects in the real world. The purpose of this paper is to extend caging-based grasping to various deformable objects. We focused on caging-based grasping of some kinds of deformable objects including pouches.

For robot hand design to reproduce the work space of the fingertip of a person, we proposed a method to measure the movement of the fingertip using optical motion capture and estimate the joint that reproduces it. For this time, trial the robot hand based on this, and make sure that the design is possible. we examine the characteristics of the joint structure of the model from the motion of the fingertip of the estimated model. As a result, compared to other robot hands, a robot hand with a large overlap of finger work spaces was designed. This is because the joints of the fingers are in a twisted relation to each other, the workspace of the fingertip has become a distribution towards the inside.

In this paper, we propose an iris mechanism robot hand using a swinging mechanism. The swing-type hand is simpler than the sliding -type hand mechanism. It can be opened and closed with a single actuator. This is a mechanism to grab the object by moving the side of the cylinder and rotating the blade around the axis. We modeled the behavior of the blade and analyzed the gap and the amount of movement of the blade.

In a versatile robotic assembly system in manufacturing, robots need to pick up, reorient and align parts with the initial pose uncertainty. It is desirable to achieve this series of manipulation with minimally necessary mechanisms and efficient strategy. In a conventional method, the series of manipulation is achieved with a single robotic hand with parallel stick fingers by utilizing environment to temporarily place parts for reorientation. For improving the efficiency of manipulation, this paper presents a method to realize picking, reorientation and alignment of shaft-shaped parts with a single robotic hand without utilizing environment, acknowledging that the parts are often handled in assembly tasks.

We propose a novel robot gripper, called Universal Vacuum Chuck (in short, UVC) based on pneumatic soft fingers covered with Universal Vacuum Gripper (in short, UVG). UVG is a vacuum adhesion cup filled with powder in a thin film.This makes it possible to combine the ability to adhere a uneven surface with UVG, and to pinch small and porous objects.We evaluated the adhesion and the gripping force of the proposed gripper. Compared to UVG, UVC improved planar adhesion force, except the uneven surface over 3 mm. Also, when gripping small objects, the gripping power was improved except small sphere.

In this paper, we propose an iris mechanism robot hand with lateral closing mechanism. The new iris mechanism robot hand is based on the iris robotic mechanism. The iris mechanism robotics hand constitution is that an iris mechanism robot hand divided in two and put on two levels to put in the lateral closing mechanism. The robot hand is able to grip the thing from itself side body and reduce quantity of operation of the robot arm than a conventional iris mechanism robot hand.

Although a tactile sensor using a camera has high spatial resolution，it has a delay time for image processing，so it is unsuitable for a task that requires a fast response．In this study, we developed a tactile image sensor that has low delay time by image processing using FPGA．We applied it to grasping force control of gripper equipped with the sensor．We confirmed that it can perform with a delay of 53 ms from force detection to grasping force change even if we use a standard 30 fps camera．

Toward long-term and non-invasive observation of nektons by an AUV without attaching any tag to them, the authors propose an autonomous detection and tracking method with a multi-beam imaging sonar. Sea turtles are set as the initial target. The method utilizes convolutional neural network (CNN) for detecting a sea turtle in sonar imagery. Surge and yaw movements of the AUV are controlled to maintain the relative distance and direction to the detected target. The proposed method was implemented in the AUV HATTORI. The AUV succeeded in tracking a sea turtle in natural condition for 270 seconds in shallow sea.

Autonomous Underwater Vehicles (AUVs) have a great potential for under-ice explorations of both the Arctic and Antarctic Oceans. Survey results of current under-ice AUVs are extremely precise, but there is a trade-off between precision and running cost and/or risk of losing AUVs. In this paper, the authors propose a variable and compact AUV MONACA (Mobility Oriented Nadir AntarctiC Adventurer), which can offer the necessary and sufficient compositions for various survey needs by the modular design. MONACA's range is smaller than conventional AUVs. However, simple and compact AUV can reduce operational risk. MONACA is expected to lower the hurdle of conducting challenging under-ice operations. Therefore, MONACA's effective operation methods were proposed together.